1 2Release 3.7.0 (XX November 2011) 3~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 43.7.0 is a feature release with many significant improvements and the 5usual collection of bug fixes. 6 7This release supports X86/Linux, AMD64/Linux, ARM/Linux, PPC32/Linux, 8PPC64/Linux, S390X/Linux, ARM/Android, X86/Darwin and AMD64/Darwin. 9Support for recent distros and toolchain components (glibc 2.14, gcc 104.6, MacOSX 10.7) has been added. 11 12* ================== PLATFORM CHANGES ================= 13 14* Support for IBM z/Architecture (s390x) running Linux. Valgrind can 15 analyse 64-bit programs running on z/Architecture. Most user space 16 instructions up to and including z10 are supported. Valgrind has 17 been tested extensively on z9, z10, and z196 machines running SLES 18 10/11, RedHat 5/6m, and Fedora. The Memcheck and Massif tools are 19 known to work well. Callgrind, Helgrind, and DRD work reasonably 20 well on z9 and later models. See README.s390 for more details. 21 22* Preliminary support for MacOSX 10.7 and XCode 4. Both 32- and 23 64-bit processes are supported. Some complex threaded applications 24 (Firefox) are observed to hang when run as 32 bit applications, 25 whereas 64-bit versions run OK. The cause is unknown. Memcheck 26 will likely report some false errors. In general, expect some rough 27 spots. This release also supports MacOSX 10.6, but drops support 28 for 10.5. 29 30* Preliminary support for Android (on ARM). Valgrind can now run 31 large applications (eg, Firefox) on (eg) a Samsung Nexus S. See 32 README.android for more details, plus instructions on how to get 33 started. 34 35* Support for the IBM Power ISA 2.06 (Power7 instructions) 36 37* General correctness and performance improvements for ARM/Linux, and, 38 by extension, ARM/Android. 39 40* Further solidification of support for SSE 4.2 in 64-bit mode. AVX 41 instruction set support is under development but is not available in 42 this release. 43 44* Support for AIX5 has been removed. 45 46* ==================== TOOL CHANGES ==================== 47 48* Memcheck: some incremental changes: 49 50 - reduction of memory use in some circumstances 51 52 - improved handling of freed memory, which in some circumstances 53 can cause detection of use-after-free that would previously have 54 been missed 55 56 - fix of a longstanding bug that could cause false negatives (missed 57 errors) in programs doing vector saturated narrowing instructions. 58 59* Helgrind: performance improvements and major memory use reductions, 60 particularly for large, long running applications which perform many 61 synchronisation (lock, unlock, etc) events. Plus many smaller 62 changes: 63 64 - display of locksets for both threads involved in a race 65 66 - general improvements in formatting/clarity of error messages 67 68 - addition of facilities and documentation regarding annotation 69 of thread safe reference counted C++ classes 70 71 - new flag --check-stack-refs=no|yes [yes], to disable race checking 72 on thread stacks (a performance hack) 73 74 - new flag --free-is-write=no|yes [no], to enable detection of races 75 where one thread accesses heap memory but another one frees it, 76 without any coordinating synchronisation event 77 78* DRD: enabled XML output; added support for delayed thread deletion 79 in order to detect races that occur close to the end of a thread 80 (--join-list-vol); fixed a memory leak triggered by repeated client 81 memory allocatation and deallocation; improved Darwin support. 82 83* exp-ptrcheck: this tool has been reduced in scope so as to improve 84 performance and remove checking that Memcheck does better. 85 Specifically, the ability to check for overruns for stack and global 86 arrays is unchanged, but the ability to check for overruns of heap 87 blocks has been removed. The tool has accordingly been renamed to 88 exp-sgcheck ("Stack and Global Array Checking"). 89 90* ==================== OTHER CHANGES ==================== 91 92* GDB server: Valgrind now has an embedded GDB server. That means it 93 is possible to control a Valgrind run from GDB, doing all the usual 94 things that GDB can do (single stepping, breakpoints, examining 95 data, etc). Tool-specific functionality is also available. For 96 example, it is possible to query the definedness state of variables 97 or memory from within GDB when running Memcheck; arbitrarily large 98 memory watchpoints are supported, etc. To use the GDB server, start 99 Valgrind with the flag --vgdb-error=0 and follow the on-screen 100 instructions. 101 102* Improved support for unfriendly self-modifying code: a new option 103 --smc-check=all-non-file is available. This adds the relevant 104 consistency checks only to code that originates in non-file-backed 105 mappings. In effect this confines the consistency checking only to 106 code that is or might be JIT generated, and avoids checks on code 107 that must have been compiled ahead of time. This significantly 108 improves performance on applications that generate code at run time. 109 110* It is now possible to build a working Valgrind using Clang-2.9 on 111 Linux. 112 113* new client requests VALGRIND_{DISABLE,ENABLE}_ERROR_REPORTING. 114 These enable and disable error reporting on a per-thread, and 115 nestable, basis. This is useful for hiding errors in particularly 116 troublesome pieces of code. The MPI wrapper library (libmpiwrap.c) 117 now uses this facility. 118 119* Added the --mod-funcname option to cg_diff. 120 121* ==================== FIXED BUGS ==================== 122 123The following bugs have been fixed or resolved. Note that "n-i-bz" 124stands for "not in bugzilla" -- that is, a bug that was reported to us 125but never got a bugzilla entry. We encourage you to file bugs in 126bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 127mailing the developers (or mailing lists) directly -- bugs that are 128not entered into bugzilla tend to get forgotten about or ignored. 129 130To see details of a given bug, visit 131https://bugs.kde.org/show_bug.cgi?id=XXXXXX 132where XXXXXX is the bug number as listed below. 133 134210935 port valgrind.h (not valgrind) to win32 to support client requests 135214223 valgrind SIGSEGV on startup gcc 4.4.1 ppc32 (G4) Ubuntu 9.10 136243404 Port to zSeries 137243935 Helgrind: incorrect handling of ANNOTATE_HAPPENS_BEFORE()/AFTER() 138247223 non-x86: Suppress warning: 'regparm' attribute directive ignored 139250101 huge "free" memory usage due to m_mallocfree.c fragmentation 140253206 Some fixes for the faultstatus testcase 141255223 capget testcase fails when running as root 142256703 xlc_dbl_u32.c testcase broken 143256726 Helgrind tests have broken inline asm 144259977 == 214223 (Valgrind segfaults doing __builtin_longjmp) 145264800 testcase compile failure on zseries 146265762 make public VEX headers compilable by G++ 3.x 147265771 assertion in jumps.c (r11523) fails with glibc-2.3 148266753 configure script does not give the user the option to not use QtCore 149266931 gen_insn_test.pl is broken 150266961 ld-linux.so.2 i?86-linux strlen issues 151266990 setns instruction causes false positive 152267020 Make directory for temporary files configurable at run-time. 153267342 == 267997 (segmentation fault on Mac OS 10.6) 154267383 Assertion 'vgPlain_strlen(dir) + vgPlain_strlen(file) + 1 < 256' failed 155267413 Assertion 'DRD_(g_threadinfo)[tid].synchr_nesting >= 1' failed. 156267488 regtest: darwin support for 64-bit build 157267552 SIGSEGV (misaligned_stack_error) with DRD, but not with other tools 158267630 Add support for IBM Power ISA 2.06 -- stage 1 159267769 == 267997 (Darwin: memcheck triggers segmentation fault) 160267819 Add client request for informing the core about reallocation 161267925 laog data structure quadratic for a single sequence of lock 162267968 drd: (vgDrd_thread_set_joinable): Assertion '0 <= (int)tid ..' failed 163267997 MacOSX: 64-bit V segfaults on launch when built with Xcode 4.0.1 164268513 missed optimizations in fold_Expr 165268619 s390x: fpr - gpr transfer facility 166268620 s390x: reconsider "long displacement" requirement 167268621 s390x: improve IR generation for XC 168268715 s390x: FLOGR is not universally available 169268792 == 267997 (valgrind seg faults on startup when compiled with Xcode 4) 170268930 s390x: MHY is not universally available 171269078 arm->IR: unhandled instruction SUB (SP minus immediate/register) 172269079 Support ptrace system call on ARM 173269144 missing "Bad option" error message 174269209 conditional load and store facility (z196) 175269354 Shift by zero on x86 can incorrectly clobber CC_NDEP 176269641 == 267997 (valgrind segfaults immediately (segmentation fault)) 177269736 s390x: minor code generation tweaks 178269778 == 272986 (valgrind.h: swap roles of VALGRIND_DO_CLIENT_REQUEST() ..) 179269863 s390x: remove unused function parameters 180269864 s390x: tweak s390_emit_load_cc 181269884 == 250101 (overhead for huge blocks exhausts space too soon) 182270082 s390x: Make sure to point the PSW address to the next address on SIGILL 183270115 s390x: rewrite some testcases 184270309 == 267997 (valgrind crash on startup) 185270320 add support for Linux FIOQSIZE ioctl() call 186270326 segfault while trying to sanitize the environment passed to execle 187270794 IBM POWER7 support patch causes regression in none/tests 188270851 IBM POWER7 fcfidus instruction causes memcheck to fail 189270856 IBM POWER7 xsnmaddadp instruction causes memcheck to fail on 32bit app 190270925 hyper-optimized strspn() in /lib64/libc-2.13.so needs fix 191270959 s390x: invalid use of R0 as base register 192271042 VSX configure check fails when it should not 193271043 Valgrind build fails with assembler error on ppc64 with binutils 2.21 194271259 s390x: fix code confusion 195271337 == 267997 (Valgrind segfaults on MacOS X) 196271385 s390x: Implement Ist_MBE 197271501 s390x: misc cleanups 198271504 s390x: promote likely and unlikely 199271579 ppc: using wrong enum type 200271615 unhandled instruction "popcnt" (arch=amd10h) 201271730 Fix bug when checking ioctls: duplicate check 202271776 s390x: provide STFLE instruction support 203271779 s390x: provide clock instructions like STCK 204271799 Darwin: ioctls without an arg report a memory error 205271820 arm: fix type confusion 206271917 pthread_cond_timedwait failure leads to not-locked false positive 207272067 s390x: fix DISP20 macro 208272615 A typo in debug output in mc_leakcheck.c 209272661 callgrind_annotate chokes when run from paths containing regex chars 210272893 amd64->IR: 0x66 0xF 0x38 0x2B 0xC1 0x66 0xF 0x7F == (closed as dup) 211272955 Unhandled syscall error for pwrite64 on ppc64 arch 212272967 make documentation build-system more robust 213272986 Fix gcc-4.6 warnings with valgrind.h 214273318 amd64->IR: 0x66 0xF 0x3A 0x61 0xC1 0x38 (missing PCMPxSTRx case) 215273318 unhandled PCMPxSTRx case: vex amd64->IR: 0x66 0xF 0x3A 0x61 0xC1 0x38 216273431 valgrind segfaults in evalCfiExpr (debuginfo.c:2039) 217273465 Callgrind: jumps.c:164 (new_jcc): Assertion '(0 <= jmp) && ...' 218273536 Build error: multiple definition of `vgDrd_pthread_cond_initializer' 219273640 ppc64-linux: unhandled syscalls setresuid(164) and setresgid(169) 220273729 == 283000 (Illegal opcode for SSE2 "roundsd" instruction) 221273778 exp-ptrcheck: unhandled sysno == 259 222274089 exp-ptrcheck: unhandled sysno == 208 223274378 s390x: Various dispatcher tweaks 224274447 WARNING: unhandled syscall: 340 225274776 amd64->IR: 0x66 0xF 0x38 0x2B 0xC5 0x66 226274784 == 267997 (valgrind ls -l results in Segmentation Fault) 227274926 valgrind does not build against linux-3 228275148 configure FAIL with glibc-2.14 229275151 Fedora 15 / glibc-2.14 'make regtest' FAIL 230275168 Make Valgrind work for MacOSX 10.7 Lion 231275212 == 275284 (lots of false positives from __memcpy_ssse3_back et al) 232275278 valgrind does not build on Linux kernel 3.0.* due to silly 233275284 Valgrind memcpy/memmove redirection stopped working in glibc 2.14/x86_64 234275308 Fix implementation for ppc64 fres instruc 235275339 s390x: fix testcase compile warnings 236275517 s390x: Provide support for CKSM instruction 237275710 s390x: get rid of redundant address mode calculation 238275815 == 247894 (Valgrind doesn't know about Linux readahead(2) syscall) 239275852 == 250101 (valgrind uses all swap space and is killed) 240276784 Add support for IBM Power ISA 2.06 -- stage 3 241276987 gdbsrv: fix tests following recent commits 242277045 Valgrind crashes with unhandled DW_OP_ opcode 0x2a 243277199 The test_isa_2_06_part1.c in none/tests/ppc64 should be a symlink 244277471 Unhandled syscall: 340 245277610 valgrind crashes in VG_(lseek)(core_fd, phdrs[idx].p_offset, ...) 246277653 ARM: support Thumb2 PLD instruction 247277663 ARM: NEON float VMUL by scalar incorrect 248277689 ARM: tests for VSTn with register post-index are broken 249277694 ARM: BLX LR instruction broken in ARM mode 250277780 ARM: VMOV.F32 (immediate) instruction is broken 251278057 fuse filesystem syscall deadlocks 252278078 Unimplemented syscall 280 on ppc32 253278349 F_GETPIPE_SZ and F_SETPIPE_SZ Linux fcntl commands 254278454 VALGRIND_STACK_DEREGISTER has wrong output type 255278502 == 275284 (Valgrind confuses memcpy() and memmove()) 256278892 gdbsrv: factorize gdb version handling, fix doc and typos 257279027 Support for MVCL and CLCL instruction 258279027 s390x: Provide support for CLCL and MVCL instructions 259279062 Remove a redundant check in the insn selector for ppc. 260279071 JDK creates PTEST with redundant REX.W prefix 261279212 gdbsrv: add monitor cmd v.info scheduler. 262279378 exp-ptrcheck: the 'impossible' happened on mkfifo call 263279698 memcheck discards valid-bits for packuswb 264279795 memcheck reports uninitialised values for mincore on amd64 265279994 Add support for IBM Power ISA 2.06 -- stage 3 266280083 mempolicy syscall check errors 267280290 vex amd64->IR: 0x66 0xF 0x38 0x28 0xC1 0x66 0xF 0x6F 268280710 s390x: config files for nightly builds 269280757 /tmp dir still used by valgrind even if TMPDIR is specified 270280965 Valgrind breaks fcntl locks when program does mmap 271281138 WARNING: unhandled syscall: 340 272281241 == 275168 (valgrind useless on Macos 10.7.1 Lion) 273281304 == 275168 (Darwin: dyld "cannot load inserted library") 274281305 == 275168 (unhandled syscall: unix:357 on Darwin 11.1) 275281468 s390x: handle do_clone and gcc clones in call traces 276281488 ARM: VFP register corruption 277281828 == 275284 (false memmove warning: "Source and destination overlap") 278281883 s390x: Fix system call wrapper for "clone". 279282105 generalise 'reclaimSuperBlock' to also reclaim splittable superblock 280282112 Unhandled instruction bytes: 0xDE 0xD9 0x9B 0xDF (fcompp) 281282238 SLES10: make check fails 282282979 strcasestr needs replacement with recent(>=2.12) glibc 283283000 vex amd64->IR: 0x66 0xF 0x3A 0xA 0xC0 0x9 0xF3 0xF 284283243 Regression in ppc64 memcheck tests 285283325 == 267997 (Darwin: V segfaults on startup when built with Xcode 4.0) 286283427 re-connect epoll_pwait syscall on ARM linux 287283600 gdbsrv: android: port vgdb.c 288283709 none/tests/faultstatus needs to account for page size 289284305 filter_gdb needs enhancement to work on ppc64 290284384 clang 3.1 -Wunused-value warnings in valgrind.h, memcheck.h 291284472 Thumb2 ROR.W encoding T2 not implemented 292284621 XML-escape process command line in XML output 293n-i-bz cachegrind/callgrind: handle CPUID information for Core iX Intel CPUs 294 that have non-power-of-2 sizes (also AMDs) 295n-i-bz don't be spooked by libraries mashed by elfhack 296n-i-bz don't be spooked by libxul.so linked with gold 297n-i-bz improved checking for VALGRIND_CHECK_MEM_IS_DEFINED 298 299(3.7.0: XX November 2011, vex rXXXX, valgrind rXXXXX). 300 301 302 303Release 3.6.1 (16 February 2011) 304~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3053.6.1 is a bug fix release. It adds support for some SSE4 306instructions that were omitted in 3.6.0 due to lack of time. Initial 307support for glibc-2.13 has been added. A number of bugs causing 308crashing or assertion failures have been fixed. 309 310The following bugs have been fixed or resolved. Note that "n-i-bz" 311stands for "not in bugzilla" -- that is, a bug that was reported to us 312but never got a bugzilla entry. We encourage you to file bugs in 313bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 314mailing the developers (or mailing lists) directly -- bugs that are 315not entered into bugzilla tend to get forgotten about or ignored. 316 317To see details of a given bug, visit 318https://bugs.kde.org/show_bug.cgi?id=XXXXXX 319where XXXXXX is the bug number as listed below. 320 321188572 Valgrind on Mac should suppress setenv() mem leak 322194402 vex amd64->IR: 0x48 0xF 0xAE 0x4 (proper FX{SAVE,RSTOR} support) 323210481 vex amd64->IR: Assertion `sz == 2 || sz == 4' failed (REX.W POPQ) 324246152 callgrind internal error after pthread_cancel on 32 Bit Linux 325250038 ppc64: Altivec LVSR and LVSL instructions fail their regtest 326254420 memory pool tracking broken 327254957 Test code failing to compile due to changes in memcheck.h 328255009 helgrind/drd: crash on chmod with invalid parameter 329255130 readdwarf3.c parse_type_DIE confused by GNAT Ada types 330255355 helgrind/drd: crash on threaded programs doing fork 331255358 == 255355 332255418 (SSE4.x) rint call compiled with ICC 333255822 --gen-suppressions can create invalid files: "too many callers [...]" 334255888 closing valgrindoutput tag outputted to log-stream on error 335255963 (SSE4.x) vex amd64->IR: 0x66 0xF 0x3A 0x9 0xDB 0x0 (ROUNDPD) 336255966 Slowness when using mempool annotations 337256387 vex x86->IR: 0xD4 0xA 0x2 0x7 (AAD and AAM) 338256600 super-optimized strcasecmp() false positive 339256669 vex amd64->IR: Unhandled LOOPNEL insn on amd64 340256968 (SSE4.x) vex amd64->IR: 0x66 0xF 0x38 0x10 0xD3 0x66 (BLENDVPx) 341257011 (SSE4.x) vex amd64->IR: 0x66 0xF 0x3A 0xE 0xFD 0xA0 (PBLENDW) 342257063 (SSE4.x) vex amd64->IR: 0x66 0xF 0x3A 0x8 0xC0 0x0 (ROUNDPS) 343257276 Missing case in memcheck --track-origins=yes 344258870 (SSE4.x) Add support for EXTRACTPS SSE 4.1 instruction 345261966 (SSE4.x) support for CRC32B and CRC32Q is lacking (also CRC32{W,L}) 346262985 VEX regression in valgrind 3.6.0 in handling PowerPC VMX 347262995 (SSE4.x) crash when trying to valgrind gcc-snapshot (PCMPxSTRx $0) 348263099 callgrind_annotate counts Ir improperly [...] 349263877 undefined coprocessor instruction on ARMv7 350265964 configure FAIL with glibc-2.13 351n-i-bz Fix compile error w/ icc-12.x in guest_arm_toIR.c 352n-i-bz Docs: fix bogus descriptions for VALGRIND_CREATE_BLOCK et al 353n-i-bz Massif: don't assert on shmat() with --pages-as-heap=yes 354n-i-bz Bug fixes and major speedups for the exp-DHAT space profiler 355n-i-bz DRD: disable --free-is-write due to implementation difficulties 356 357(3.6.1: 16 February 2011, vex r2103, valgrind r11561). 358 359 360 361Release 3.6.0 (21 October 2010) 362~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3633.6.0 is a feature release with many significant improvements and the 364usual collection of bug fixes. 365 366This release supports X86/Linux, AMD64/Linux, ARM/Linux, PPC32/Linux, 367PPC64/Linux, X86/Darwin and AMD64/Darwin. Support for recent distros 368and toolchain components (glibc 2.12, gcc 4.5, OSX 10.6) has been added. 369 370 ------------------------- 371 372Here are some highlights. Details are shown further down: 373 374* Support for ARM/Linux. 375 376* Support for recent Linux distros: Ubuntu 10.10 and Fedora 14. 377 378* Support for Mac OS X 10.6, both 32- and 64-bit executables. 379 380* Support for the SSE4.2 instruction set. 381 382* Enhancements to the Callgrind profiler, including the ability to 383 handle CPUs with three levels of cache. 384 385* A new experimental heap profiler, DHAT. 386 387* A huge number of bug fixes and small enhancements. 388 389 ------------------------- 390 391Here are details of the above changes, together with descriptions of 392many other changes, and a list of fixed bugs. 393 394* ================== PLATFORM CHANGES ================= 395 396* Support for ARM/Linux. Valgrind now runs on ARMv7 capable CPUs 397 running Linux. It is known to work on Ubuntu 10.04, Ubuntu 10.10, 398 and Maemo 5, so you can run Valgrind on your Nokia N900 if you want. 399 400 This requires a CPU capable of running the ARMv7-A instruction set 401 (Cortex A5, A8 and A9). Valgrind provides fairly complete coverage 402 of the user space instruction set, including ARM and Thumb integer 403 code, VFPv3, NEON and V6 media instructions. The Memcheck, 404 Cachegrind and Massif tools work properly; other tools work to 405 varying degrees. 406 407* Support for recent Linux distros (Ubuntu 10.10 and Fedora 14), along 408 with support for recent releases of the underlying toolchain 409 components, notably gcc-4.5 and glibc-2.12. 410 411* Support for Mac OS X 10.6, both 32- and 64-bit executables. 64-bit 412 support also works much better on OS X 10.5, and is as solid as 413 32-bit support now. 414 415* Support for the SSE4.2 instruction set. SSE4.2 is supported in 416 64-bit mode. In 32-bit mode, support is only available up to and 417 including SSSE3. Some exceptions: SSE4.2 AES instructions are not 418 supported in 64-bit mode, and 32-bit mode does in fact support the 419 bare minimum SSE4 instructions to needed to run programs on Mac OS X 420 10.6 on 32-bit targets. 421 422* Support for IBM POWER6 cpus has been improved. The Power ISA up to 423 and including version 2.05 is supported. 424 425* ==================== TOOL CHANGES ==================== 426 427* Cachegrind has a new processing script, cg_diff, which finds the 428 difference between two profiles. It's very useful for evaluating 429 the performance effects of a change in a program. 430 431 Related to this change, the meaning of cg_annotate's (rarely-used) 432 --threshold option has changed; this is unlikely to affect many 433 people, if you do use it please see the user manual for details. 434 435* Callgrind now can do branch prediction simulation, similar to 436 Cachegrind. In addition, it optionally can count the number of 437 executed global bus events. Both can be used for a better 438 approximation of a "Cycle Estimation" as derived event (you need to 439 update the event formula in KCachegrind yourself). 440 441* Cachegrind and Callgrind now refer to the LL (last-level) cache 442 rather than the L2 cache. This is to accommodate machines with 443 three levels of caches -- if Cachegrind/Callgrind auto-detects the 444 cache configuration of such a machine it will run the simulation as 445 if the L2 cache isn't present. This means the results are less 446 likely to match the true result for the machine, but 447 Cachegrind/Callgrind's results are already only approximate, and 448 should not be considered authoritative. The results are still 449 useful for giving a general idea about a program's locality. 450 451* Massif has a new option, --pages-as-heap, which is disabled by 452 default. When enabled, instead of tracking allocations at the level 453 of heap blocks (as allocated with malloc/new/new[]), it instead 454 tracks memory allocations at the level of memory pages (as mapped by 455 mmap, brk, etc). Each mapped page is treated as its own block. 456 Interpreting the page-level output is harder than the heap-level 457 output, but this option is useful if you want to account for every 458 byte of memory used by a program. 459 460* DRD has two new command-line options: --free-is-write and 461 --trace-alloc. The former allows to detect reading from already freed 462 memory, and the latter allows tracing of all memory allocations and 463 deallocations. 464 465* DRD has several new annotations. Custom barrier implementations can 466 now be annotated, as well as benign races on static variables. 467 468* DRD's happens before / happens after annotations have been made more 469 powerful, so that they can now also be used to annotate e.g. a smart 470 pointer implementation. 471 472* Helgrind's annotation set has also been drastically improved, so as 473 to provide to users a general set of annotations to describe locks, 474 semaphores, barriers and condition variables. Annotations to 475 describe thread-safe reference counted heap objects have also been 476 added. 477 478* Memcheck has a new command-line option, --show-possibly-lost, which 479 is enabled by default. When disabled, the leak detector will not 480 show possibly-lost blocks. 481 482* A new experimental heap profiler, DHAT (Dynamic Heap Analysis Tool), 483 has been added. DHAT keeps track of allocated heap blocks, and also 484 inspects every memory reference to see which block (if any) is being 485 accessed. This gives a lot of insight into block lifetimes, 486 utilisation, turnover, liveness, and the location of hot and cold 487 fields. You can use DHAT to do hot-field profiling. 488 489* ==================== OTHER CHANGES ==================== 490 491* Improved support for unfriendly self-modifying code: the extra 492 overhead incurred by --smc-check=all has been reduced by 493 approximately a factor of 5 as compared with 3.5.0. 494 495* Ability to show directory names for source files in error messages. 496 This is combined with a flexible mechanism for specifying which 497 parts of the paths should be shown. This is enabled by the new flag 498 --fullpath-after. 499 500* A new flag, --require-text-symbol, which will stop the run if a 501 specified symbol is not found it a given shared object when it is 502 loaded into the process. This makes advanced working with function 503 intercepting and wrapping safer and more reliable. 504 505* Improved support for the Valkyrie GUI, version 2.0.0. GUI output 506 and control of Valgrind is now available for the tools Memcheck and 507 Helgrind. XML output from Valgrind is available for Memcheck, 508 Helgrind and exp-Ptrcheck. 509 510* More reliable stack unwinding on amd64-linux, particularly in the 511 presence of function wrappers, and with gcc-4.5 compiled code. 512 513* Modest scalability (performance improvements) for massive 514 long-running applications, particularly for those with huge amounts 515 of code. 516 517* Support for analyzing programs running under Wine with has been 518 improved. The header files <valgrind/valgrind.h>, 519 <valgrind/memcheck.h> and <valgrind/drd.h> can now be used in 520 Windows-programs compiled with MinGW or one of the Microsoft Visual 521 Studio compilers. 522 523* A rare but serious error in the 64-bit x86 CPU simulation was fixed. 524 The 32-bit simulator was not affected. This did not occur often, 525 but when it did would usually crash the program under test. 526 Bug 245925. 527 528* A large number of bugs were fixed. These are shown below. 529 530* A number of bugs were investigated, and were candidates for fixing, 531 but are not fixed in 3.6.0, due to lack of developer time. They may 532 get fixed in later releases. They are: 533 534 194402 vex amd64->IR: 0x48 0xF 0xAE 0x4 0x24 0x49 (FXSAVE64) 535 212419 false positive "lock order violated" (A+B vs A) 536 213685 Undefined value propagates past dependency breaking instruction 537 216837 Incorrect instrumentation of NSOperationQueue on Darwin 538 237920 valgrind segfault on fork failure 539 242137 support for code compiled by LLVM-2.8 540 242423 Another unknown Intel cache config value 541 243232 Inconsistent Lock Orderings report with trylock 542 243483 ppc: callgrind triggers VEX assertion failure 543 243935 Helgrind: implementation of ANNOTATE_HAPPENS_BEFORE() is wrong 544 244677 Helgrind crash hg_main.c:616 (map_threads_lookup): Assertion 545 'thr' failed. 546 246152 callgrind internal error after pthread_cancel on 32 Bit Linux 547 249435 Analyzing wine programs with callgrind triggers a crash 548 250038 ppc64: Altivec lvsr and lvsl instructions fail their regtest 549 250065 Handling large allocations 550 250101 huge "free" memory usage due to m_mallocfree.c 551 "superblocks fragmentation" 552 251569 vex amd64->IR: 0xF 0x1 0xF9 0x8B 0x4C 0x24 (RDTSCP) 553 252091 Callgrind on ARM does not detect function returns correctly 554 252600 [PATCH] Allow lhs to be a pointer for shl/shr 555 254420 memory pool tracking broken 556 n-i-bz support for adding symbols for JIT generated code 557 558 559The following bugs have been fixed or resolved. Note that "n-i-bz" 560stands for "not in bugzilla" -- that is, a bug that was reported to us 561but never got a bugzilla entry. We encourage you to file bugs in 562bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 563mailing the developers (or mailing lists) directly -- bugs that are 564not entered into bugzilla tend to get forgotten about or ignored. 565 566To see details of a given bug, visit 567https://bugs.kde.org/show_bug.cgi?id=XXXXXX 568where XXXXXX is the bug number as listed below. 569 570135264 dcbzl instruction missing 571142688 == 250799 572153699 Valgrind should report unaligned reads with movdqa 573180217 == 212335 574190429 Valgrind reports lost of errors in ld.so 575 with x86_64 2.9.90 glibc 576197266 valgrind appears to choke on the xmms instruction 577 "roundsd" on x86_64 578197988 Crash when demangling very large symbol names 579202315 unhandled syscall: 332 (inotify_init1) 580203256 Add page-level profiling to Massif 581205093 dsymutil=yes needs quotes, locking (partial fix) 582205241 Snow Leopard 10.6 support (partial fix) 583206600 Leak checker fails to upgrade indirect blocks when their 584 parent becomes reachable 585210935 port valgrind.h (not valgrind) to win32 so apps run under 586 wine can make client requests 587211410 vex amd64->IR: 0x15 0xFF 0xFF 0x0 0x0 0x89 588 within Linux ip-stack checksum functions 589212335 unhandled instruction bytes: 0xF3 0xF 0xBD 0xC0 590 (lzcnt %eax,%eax) 591213685 Undefined value propagates past dependency breaking instruction 592 (partial fix) 593215914 Valgrind inserts bogus empty environment variable 594217863 == 197988 595219538 adjtimex syscall wrapper wrong in readonly adjtime mode 596222545 shmat fails under valgind on some arm targets 597222560 ARM NEON support 598230407 == 202315 599231076 == 202315 600232509 Docs build fails with formatting inside <title></title> elements 601232793 == 202315 602235642 [PATCH] syswrap-linux.c: support evdev EVIOCG* ioctls 603236546 vex x86->IR: 0x66 0xF 0x3A 0xA 604237202 vex amd64->IR: 0xF3 0xF 0xB8 0xC0 0x49 0x3B 605237371 better support for VALGRIND_MALLOCLIKE_BLOCK 606237485 symlink (syscall 57) is not supported on Mac OS 607237723 sysno == 101 exp-ptrcheck: the 'impossible' happened: 608 unhandled syscall 609238208 is_just_below_ESP doesn't take into account red-zone 610238345 valgrind passes wrong $0 when executing a shell script 611238679 mq_timedreceive syscall doesn't flag the reception buffer 612 as "defined" 613238696 fcntl command F_DUPFD_CLOEXEC not supported 614238713 unhandled instruction bytes: 0x66 0xF 0x29 0xC6 615238713 unhandled instruction bytes: 0x66 0xF 0x29 0xC6 616238745 3.5.0 Make fails on PPC Altivec opcodes, though configure 617 says "Altivec off" 618239992 vex amd64->IR: 0x48 0xF 0xC4 0xC1 0x0 0x48 619240488 == 197988 620240639 == 212335 621241377 == 236546 622241903 == 202315 623241920 == 212335 624242606 unhandled syscall: setegid (in Ptrcheck) 625242814 Helgrind "Impossible has happened" during 626 QApplication::initInstance(); 627243064 Valgrind attempting to read debug information from iso 628243270 Make stack unwinding in Valgrind wrappers more reliable 629243884 exp-ptrcheck: the 'impossible happened: unhandled syscall 630 sysno = 277 (mq_open) 631244009 exp-ptrcheck unknown syscalls in analyzing lighttpd 632244493 ARM VFP d16-d31 registers support 633244670 add support for audit_session_self syscall on Mac OS 10.6 634244921 The xml report of helgrind tool is not well format 635244923 In the xml report file, the <preamble> not escape the 636 xml char, eg '<','&','>' 637245535 print full path names in plain text reports 638245925 x86-64 red zone handling problem 639246258 Valgrind not catching integer underruns + new [] s 640246311 reg/reg cmpxchg doesn't work on amd64 641246549 unhandled syscall unix:277 while testing 32-bit Darwin app 642246888 Improve Makefile.vex.am 643247510 [OS X 10.6] Memcheck reports unaddressable bytes passed 644 to [f]chmod_extended 645247526 IBM POWER6 (ISA 2.05) support is incomplete 646247561 Some leak testcases fails due to reachable addresses in 647 caller save regs 648247875 sizeofIRType to handle Ity_I128 649247894 [PATCH] unhandled syscall sys_readahead 650247980 Doesn't honor CFLAGS passed to configure 651248373 darwin10.supp is empty in the trunk 652248822 Linux FIBMAP ioctl has int parameter instead of long 653248893 [PATCH] make readdwarf.c big endianess safe to enable 654 unwinding on big endian systems 655249224 Syscall 336 not supported (SYS_proc_info) 656249359 == 245535 657249775 Incorrect scheme for detecting NEON capabilities of host CPU 658249943 jni JVM init fails when using valgrind 659249991 Valgrind incorrectly declares AESKEYGENASSIST support 660 since VEX r2011 661249996 linux/arm: unhandled syscall: 181 (__NR_pwrite64) 662250799 frexp$fenv_access_off function generates SIGILL 663250998 vex x86->IR: unhandled instruction bytes: 0x66 0x66 0x66 0x2E 664251251 support pclmulqdq insn 665251362 valgrind: ARM: attach to debugger either fails or provokes 666 kernel oops 667251674 Unhandled syscall 294 668251818 == 254550 669 670254257 Add support for debugfiles found by build-id 671254550 [PATCH] Implement DW_ATE_UTF (DWARF4) 672254646 Wrapped functions cause stack misalignment on OS X 673 (and possibly Linux) 674254556 ARM: valgrinding anything fails with SIGSEGV for 0xFFFF0FA0 675 676(3.6.0: 21 October 2010, vex r2068, valgrind r11471). 677 678 679 680Release 3.5.0 (19 August 2009) 681~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 6823.5.0 is a feature release with many significant improvements and the 683usual collection of bug fixes. The main improvement is that Valgrind 684now works on Mac OS X. 685 686This release supports X86/Linux, AMD64/Linux, PPC32/Linux, PPC64/Linux 687and X86/Darwin. Support for recent distros and toolchain components 688(glibc 2.10, gcc 4.5) has been added. 689 690 ------------------------- 691 692Here is a short summary of the changes. Details are shown further 693down: 694 695* Support for Mac OS X (10.5.x). 696 697* Improvements and simplifications to Memcheck's leak checker. 698 699* Clarification and simplifications in various aspects of Valgrind's 700 text output. 701 702* XML output for Helgrind and Ptrcheck. 703 704* Performance and stability improvements for Helgrind and DRD. 705 706* Genuinely atomic support for x86/amd64/ppc atomic instructions. 707 708* A new experimental tool, BBV, useful for computer architecture 709 research. 710 711* Improved Wine support, including ability to read Windows PDB 712 debuginfo. 713 714 ------------------------- 715 716Here are details of the above changes, followed by descriptions of 717many other minor changes, and a list of fixed bugs. 718 719 720* Valgrind now runs on Mac OS X. (Note that Mac OS X is sometimes 721 called "Darwin" because that is the name of the OS core, which is the 722 level that Valgrind works at.) 723 724 Supported systems: 725 726 - It requires OS 10.5.x (Leopard). Porting to 10.4.x is not planned 727 because it would require work and 10.4 is only becoming less common. 728 729 - 32-bit programs on x86 and AMD64 (a.k.a x86-64) machines are supported 730 fairly well. For 10.5.x, 32-bit programs are the default even on 731 64-bit machines, so it handles most current programs. 732 733 - 64-bit programs on x86 and AMD64 (a.k.a x86-64) machines are not 734 officially supported, but simple programs at least will probably work. 735 However, start-up is slow. 736 737 - PowerPC machines are not supported. 738 739 Things that don't work: 740 741 - The Ptrcheck tool. 742 743 - Objective-C garbage collection. 744 745 - --db-attach=yes. 746 747 - If you have Rogue Amoeba's "Instant Hijack" program installed, 748 Valgrind will fail with a SIGTRAP at start-up. See 749 https://bugs.kde.org/show_bug.cgi?id=193917 for details and a 750 simple work-around. 751 752 Usage notes: 753 754 - You will likely find --dsymutil=yes a useful option, as error 755 messages may be imprecise without it. 756 757 - Mac OS X support is new and therefore will be less robust than the 758 Linux support. Please report any bugs you find. 759 760 - Threaded programs may run more slowly than on Linux. 761 762 Many thanks to Greg Parker for developing this port over several years. 763 764 765* Memcheck's leak checker has been improved. 766 767 - The results for --leak-check=summary now match the summary results 768 for --leak-check=full. Previously they could differ because 769 --leak-check=summary counted "indirectly lost" blocks and 770 "suppressed" blocks as "definitely lost". 771 772 - Blocks that are only reachable via at least one interior-pointer, 773 but are directly pointed to by a start-pointer, were previously 774 marked as "still reachable". They are now correctly marked as 775 "possibly lost". 776 777 - The default value for the --leak-resolution option has been 778 changed from "low" to "high". In general, this means that more 779 leak reports will be produced, but each leak report will describe 780 fewer leaked blocks. 781 782 - With --leak-check=full, "definitely lost" and "possibly lost" 783 leaks are now considered as proper errors, ie. they are counted 784 for the "ERROR SUMMARY" and affect the behaviour of 785 --error-exitcode. These leaks are not counted as errors if 786 --leak-check=summary is specified, however. 787 788 - Documentation for the leak checker has been improved. 789 790 791* Various aspects of Valgrind's text output have changed. 792 793 - Valgrind's start-up message has changed. It is shorter but also 794 includes the command being run, which makes it easier to use 795 --trace-children=yes. An example: 796 797 - Valgrind's shut-down messages have also changed. This is most 798 noticeable with Memcheck, where the leak summary now occurs before 799 the error summary. This change was necessary to allow leaks to be 800 counted as proper errors (see the description of the leak checker 801 changes above for more details). This was also necessary to fix a 802 longstanding bug in which uses of suppressions against leaks were 803 not "counted", leading to difficulties in maintaining suppression 804 files (see https://bugs.kde.org/show_bug.cgi?id=186790). 805 806 - Behavior of -v has changed. In previous versions, -v printed out 807 a mixture of marginally-user-useful information, and tool/core 808 statistics. The statistics printing has now been moved to its own 809 flag, --stats=yes. This means -v is less verbose and more likely 810 to convey useful end-user information. 811 812 - The format of some (non-XML) stack trace entries has changed a 813 little. Previously there were six possible forms: 814 815 0x80483BF: really (a.c:20) 816 0x80483BF: really (in /foo/a.out) 817 0x80483BF: really 818 0x80483BF: (within /foo/a.out) 819 0x80483BF: ??? (a.c:20) 820 0x80483BF: ??? 821 822 The third and fourth of these forms have been made more consistent 823 with the others. The six possible forms are now: 824 825 0x80483BF: really (a.c:20) 826 0x80483BF: really (in /foo/a.out) 827 0x80483BF: really (in ???) 828 0x80483BF: ??? (in /foo/a.out) 829 0x80483BF: ??? (a.c:20) 830 0x80483BF: ??? 831 832 Stack traces produced when --xml=yes is specified are different 833 and unchanged. 834 835 836* Helgrind and Ptrcheck now support XML output, so they can be used 837 from GUI tools. Also, the XML output mechanism has been 838 overhauled. 839 840 - The XML format has been overhauled and generalised, so it is more 841 suitable for error reporting tools in general. The Memcheck 842 specific aspects of it have been removed. The new format, which 843 is an evolution of the old format, is described in 844 docs/internals/xml-output-protocol4.txt. 845 846 - Memcheck has been updated to use the new format. 847 848 - Helgrind and Ptrcheck are now able to emit output in this format. 849 850 - The XML output mechanism has been overhauled. XML is now output 851 to its own file descriptor, which means that: 852 853 * Valgrind can output text and XML independently. 854 855 * The longstanding problem of XML output being corrupted by 856 unexpected un-tagged text messages is solved. 857 858 As before, the destination for text output is specified using 859 --log-file=, --log-fd= or --log-socket=. 860 861 As before, XML output for a tool is enabled using --xml=yes. 862 863 Because there's a new XML output channel, the XML output 864 destination is now specified by --xml-file=, --xml-fd= or 865 --xml-socket=. 866 867 Initial feedback has shown this causes some confusion. To 868 clarify, the two envisaged usage scenarios are: 869 870 (1) Normal text output. In this case, do not specify --xml=yes 871 nor any of --xml-file=, --xml-fd= or --xml-socket=. 872 873 (2) XML output. In this case, specify --xml=yes, and one of 874 --xml-file=, --xml-fd= or --xml-socket= to select the XML 875 destination, one of --log-file=, --log-fd= or --log-socket= 876 to select the destination for any remaining text messages, 877 and, importantly, -q. 878 879 -q makes Valgrind completely silent on the text channel, 880 except in the case of critical failures, such as Valgrind 881 itself segfaulting, or failing to read debugging information. 882 Hence, in this scenario, it suffices to check whether or not 883 any output appeared on the text channel. If yes, then it is 884 likely to be a critical error which should be brought to the 885 attention of the user. If no (the text channel produced no 886 output) then it can be assumed that the run was successful. 887 888 This allows GUIs to make the critical distinction they need to 889 make (did the run fail or not?) without having to search or 890 filter the text output channel in any way. 891 892 It is also recommended to use --child-silent-after-fork=yes in 893 scenario (2). 894 895 896* Improvements and changes in Helgrind: 897 898 - XML output, as described above 899 900 - Checks for consistent association between pthread condition 901 variables and their associated mutexes are now performed. 902 903 - pthread_spinlock functions are supported. 904 905 - Modest performance improvements. 906 907 - Initial (skeletal) support for describing the behaviour of 908 non-POSIX synchronisation objects through ThreadSanitizer 909 compatible ANNOTATE_* macros. 910 911 - More controllable tradeoffs between performance and the level of 912 detail of "previous" accesses in a race. There are now three 913 settings: 914 915 * --history-level=full. This is the default, and was also the 916 default in 3.4.x. It shows both stacks involved in a race, but 917 requires a lot of memory and can be very slow in programs that 918 do many inter-thread synchronisation events. 919 920 * --history-level=none. This only shows the later stack involved 921 in a race. This can be much faster than --history-level=full, 922 but makes it much more difficult to find the other access 923 involved in the race. 924 925 The new intermediate setting is 926 927 * --history-level=approx 928 929 For the earlier (other) access, two stacks are presented. The 930 earlier access is guaranteed to be somewhere in between the two 931 program points denoted by those stacks. This is not as useful 932 as showing the exact stack for the previous access (as per 933 --history-level=full), but it is better than nothing, and it's 934 almost as fast as --history-level=none. 935 936 937* New features and improvements in DRD: 938 939 - The error messages printed by DRD are now easier to interpret. 940 Instead of using two different numbers to identify each thread 941 (Valgrind thread ID and DRD thread ID), DRD does now identify 942 threads via a single number (the DRD thread ID). Furthermore 943 "first observed at" information is now printed for all error 944 messages related to synchronization objects. 945 946 - Added support for named semaphores (sem_open() and sem_close()). 947 948 - Race conditions between pthread_barrier_wait() and 949 pthread_barrier_destroy() calls are now reported. 950 951 - Added support for custom allocators through the macros 952 VALGRIND_MALLOCLIKE_BLOCK() VALGRIND_FREELIKE_BLOCK() (defined in 953 in <valgrind/valgrind.h>). An alternative for these two macros is 954 the new client request VG_USERREQ__DRD_CLEAN_MEMORY (defined in 955 <valgrind/drd.h>). 956 957 - Added support for annotating non-POSIX synchronization objects 958 through several new ANNOTATE_*() macros. 959 960 - OpenMP: added support for the OpenMP runtime (libgomp) included 961 with gcc versions 4.3.0 and 4.4.0. 962 963 - Faster operation. 964 965 - Added two new command-line options (--first-race-only and 966 --segment-merging-interval). 967 968 969* Genuinely atomic support for x86/amd64/ppc atomic instructions 970 971 Valgrind will now preserve (memory-access) atomicity of LOCK- 972 prefixed x86/amd64 instructions, and any others implying a global 973 bus lock. Ditto for PowerPC l{w,d}arx/st{w,d}cx. instructions. 974 975 This means that Valgrinded processes will "play nicely" in 976 situations where communication with other processes, or the kernel, 977 is done through shared memory and coordinated with such atomic 978 instructions. Prior to this change, such arrangements usually 979 resulted in hangs, races or other synchronisation failures, because 980 Valgrind did not honour atomicity of such instructions. 981 982 983* A new experimental tool, BBV, has been added. BBV generates basic 984 block vectors for use with the SimPoint analysis tool, which allows 985 a program's overall behaviour to be approximated by running only a 986 fraction of it. This is useful for computer architecture 987 researchers. You can run BBV by specifying --tool=exp-bbv (the 988 "exp-" prefix is short for "experimental"). BBV was written by 989 Vince Weaver. 990 991 992* Modestly improved support for running Windows applications under 993 Wine. In particular, initial support for reading Windows .PDB debug 994 information has been added. 995 996 997* A new Memcheck client request VALGRIND_COUNT_LEAK_BLOCKS has been 998 added. It is similar to VALGRIND_COUNT_LEAKS but counts blocks 999 instead of bytes. 1000 1001 1002* The Valgrind client requests VALGRIND_PRINTF and 1003 VALGRIND_PRINTF_BACKTRACE have been changed slightly. Previously, 1004 the string was always printed immediately on its own line. Now, the 1005 string will be added to a buffer but not printed until a newline is 1006 encountered, or other Valgrind output is printed (note that for 1007 VALGRIND_PRINTF_BACKTRACE, the back-trace itself is considered 1008 "other Valgrind output"). This allows you to use multiple 1009 VALGRIND_PRINTF calls to build up a single output line, and also to 1010 print multiple output lines with a single request (by embedding 1011 multiple newlines in the string). 1012 1013 1014* The graphs drawn by Massif's ms_print program have changed slightly: 1015 1016 - The half-height chars '.' and ',' are no longer drawn, because 1017 they are confusing. The --y option can be used if the default 1018 y-resolution is not high enough. 1019 1020 - Horizontal lines are now drawn after the top of a snapshot if 1021 there is a gap until the next snapshot. This makes it clear that 1022 the memory usage has not dropped to zero between snapshots. 1023 1024 1025* Something that happened in 3.4.0, but wasn't clearly announced: the 1026 option --read-var-info=yes can be used by some tools (Memcheck, 1027 Helgrind and DRD). When enabled, it causes Valgrind to read DWARF3 1028 variable type and location information. This makes those tools 1029 start up more slowly and increases memory consumption, but 1030 descriptions of data addresses in error messages become more 1031 detailed. 1032 1033 1034* exp-Omega, an experimental instantaneous leak-detecting tool, was 1035 disabled in 3.4.0 due to a lack of interest and maintenance, 1036 although the source code was still in the distribution. The source 1037 code has now been removed from the distribution. For anyone 1038 interested, the removal occurred in SVN revision r10247. 1039 1040 1041* Some changes have been made to the build system. 1042 1043 - VEX/ is now integrated properly into the build system. This means 1044 that dependency tracking within VEX/ now works properly, "make 1045 install" will work without requiring "make" before it, and 1046 parallel builds (ie. 'make -j') now work (previously a 1047 .NOTPARALLEL directive was used to serialize builds, ie. 'make -j' 1048 was effectively ignored). 1049 1050 - The --with-vex configure option has been removed. It was of 1051 little use and removing it simplified the build system. 1052 1053 - The location of some install files has changed. This should not 1054 affect most users. Those who might be affected: 1055 1056 * For people who use Valgrind with MPI programs, the installed 1057 libmpiwrap.so library has moved from 1058 $(INSTALL)/<platform>/libmpiwrap.so to 1059 $(INSTALL)/libmpiwrap-<platform>.so. 1060 1061 * For people who distribute standalone Valgrind tools, the 1062 installed libraries such as $(INSTALL)/<platform>/libcoregrind.a 1063 have moved to $(INSTALL)/libcoregrind-<platform>.a. 1064 1065 These changes simplify the build system. 1066 1067 - Previously, all the distributed suppression (*.supp) files were 1068 installed. Now, only default.supp is installed. This should not 1069 affect users as the other installed suppression files were not 1070 read; the fact that they were installed was a mistake. 1071 1072 1073* KNOWN LIMITATIONS: 1074 1075 - Memcheck is unusable with the Intel compiler suite version 11.1, 1076 when it generates code for SSE2-and-above capable targets. This 1077 is because of icc's use of highly optimised inlined strlen 1078 implementations. It causes Memcheck to report huge numbers of 1079 false errors even in simple programs. Helgrind and DRD may also 1080 have problems. 1081 1082 Versions 11.0 and earlier may be OK, but this has not been 1083 properly tested. 1084 1085 1086The following bugs have been fixed or resolved. Note that "n-i-bz" 1087stands for "not in bugzilla" -- that is, a bug that was reported to us 1088but never got a bugzilla entry. We encourage you to file bugs in 1089bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 1090mailing the developers (or mailing lists) directly -- bugs that are 1091not entered into bugzilla tend to get forgotten about or ignored. 1092 1093To see details of a given bug, visit 1094https://bugs.kde.org/show_bug.cgi?id=XXXXXX 1095where XXXXXX is the bug number as listed below. 1096 109784303 How about a LockCheck tool? 109891633 dereference of null ptr in vgPlain_st_basetype 109997452 Valgrind doesn't report any pthreads problems 1100100628 leak-check gets assertion failure when using 1101 VALGRIND_MALLOCLIKE_BLOCK on malloc()ed memory 1102108528 NPTL pthread cleanup handlers not called 1103110126 Valgrind 2.4.1 configure.in tramples CFLAGS 1104110128 mallinfo is not implemented... 1105110770 VEX: Generated files not always updated when making valgrind 1106111102 Memcheck: problems with large (memory footprint) applications 1107115673 Vex's decoder should never assert 1108117564 False positive: Syscall param clone(child_tidptr) contains 1109 uninitialised byte(s) 1110119404 executing ssh from inside valgrind fails 1111133679 Callgrind does not write path names to sources with dwarf debug 1112 info 1113135847 configure.in problem with non gnu compilers (and possible fix) 1114136154 threads.c:273 (vgCallgrind_post_signal): Assertion 1115 '*(vgCallgrind_current_fn_stack.top) == 0' failed. 1116136230 memcheck reports "possibly lost", should be "still reachable" 1117137073 NULL arg to MALLOCLIKE_BLOCK causes crash 1118137904 Valgrind reports a memory leak when using POSIX threads, 1119 while it shouldn't 1120139076 valgrind VT_GETSTATE error 1121142228 complaint of elf_dynamic_do_rela in trivial usage 1122145347 spurious warning with USBDEVFS_REAPURB 1123148441 (wine) can't find memory leak in Wine, win32 binary 1124 executable file. 1125148742 Leak-check fails assert on exit 1126149878 add (proper) check for calloc integer overflow 1127150606 Call graph is broken when using callgrind control 1128152393 leak errors produce an exit code of 0. I need some way to 1129 cause leak errors to result in a nonzero exit code. 1130157154 documentation (leak-resolution doc speaks about num-callers 1131 def=4) + what is a loss record 1132159501 incorrect handling of ALSA ioctls 1133162020 Valgrinding an empty/zero-byte file crashes valgrind 1134162482 ppc: Valgrind crashes while reading stabs information 1135162718 x86: avoid segment selector 0 in sys_set_thread_area() 1136163253 (wine) canonicaliseSymtab forgot some fields in DiSym 1137163560 VEX/test_main.c is missing from valgrind-3.3.1 1138164353 malloc_usable_size() doesn't return a usable size 1139165468 Inconsistent formatting in memcheck manual -- please fix 1140169505 main.c:286 (endOfInstr): 1141 Assertion 'ii->cost_offset == *cost_offset' failed 1142177206 Generate default.supp during compile instead of configure 1143177209 Configure valt_load_address based on arch+os 1144177305 eventfd / syscall 323 patch lost 1145179731 Tests fail to build because of inlining of non-local asm labels 1146181394 helgrind: libhb_core.c:3762 (msm_write): Assertion 1147 'ordxx == POrd_EQ || ordxx == POrd_LT' failed. 1148181594 Bogus warning for empty text segment 1149181707 dwarf doesn't require enumerations to have name 1150185038 exp-ptrcheck: "unhandled syscall: 285" (fallocate) on x86_64 1151185050 exp-ptrcheck: sg_main.c:727 (add_block_to_GlobalTree): 1152 Assertion '!already_present' failed. 1153185359 exp-ptrcheck: unhandled syscall getresuid() 1154185794 "WARNING: unhandled syscall: 285" (fallocate) on x86_64 1155185816 Valgrind is unable to handle debug info for files with split 1156 debug info that are prelinked afterwards 1157185980 [darwin] unhandled syscall: sem_open 1158186238 bbToIR_AMD64: disInstr miscalculated next %rip 1159186507 exp-ptrcheck unhandled syscalls prctl, etc. 1160186790 Suppression pattern used for leaks are not reported 1161186796 Symbols with length>200 in suppression files are ignored 1162187048 drd: mutex PTHREAD_PROCESS_SHARED attribute missinterpretation 1163187416 exp-ptrcheck: support for __NR_{setregid,setreuid,setresuid} 1164188038 helgrind: hg_main.c:926: mk_SHVAL_fail: the 'impossible' happened 1165188046 bashisms in the configure script 1166188127 amd64->IR: unhandled instruction bytes: 0xF0 0xF 0xB0 0xA 1167188161 memcheck: --track-origins=yes asserts "mc_machine.c:672 1168 (get_otrack_shadow_offset_wrk): the 'impossible' happened." 1169188248 helgrind: pthread_cleanup_push, pthread_rwlock_unlock, 1170 assertion fail "!lock->heldBy" 1171188427 Add support for epoll_create1 (with patch) 1172188530 Support for SIOCGSTAMPNS 1173188560 Include valgrind.spec in the tarball 1174188572 Valgrind on Mac should suppress setenv() mem leak 1175189054 Valgrind fails to build because of duplicate non-local asm labels 1176189737 vex amd64->IR: unhandled instruction bytes: 0xAC 1177189762 epoll_create syscall not handled (--tool=exp-ptrcheck) 1178189763 drd assertion failure: s_threadinfo[tid].is_recording 1179190219 unhandled syscall: 328 (x86-linux) 1180190391 dup of 181394; see above 1181190429 Valgrind reports lots of errors in ld.so with x86_64 2.9.90 glibc 1182190820 No debug information on powerpc-linux 1183191095 PATCH: Improve usbdevfs ioctl handling 1184191182 memcheck: VALGRIND_LEAK_CHECK quadratic when big nr of chunks 1185 or big nr of errors 1186191189 --xml=yes should obey --gen-suppressions=all 1187191192 syslog() needs a suppression on macosx 1188191271 DARWIN: WARNING: unhandled syscall: 33554697 a.k.a.: 265 1189191761 getrlimit on MacOSX 1190191992 multiple --fn-skip only works sometimes; dependent on order 1191192634 V. reports "aspacem sync_check_mapping_callback: 1192 segment mismatch" on Darwin 1193192954 __extension__ missing on 2 client requests 1194194429 Crash at start-up with glibc-2.10.1 and linux-2.6.29 1195194474 "INSTALL" file has different build instructions than "README" 1196194671 Unhandled syscall (sem_wait?) from mac valgrind 1197195069 memcheck: reports leak (memory still reachable) for 1198 printf("%d', x) 1199195169 drd: (vgDrd_barrier_post_wait): 1200 Assertion 'r->sg[p->post_iteration]' failed. 1201195268 valgrind --log-file doesn't accept ~/... 1202195838 VEX abort: LibVEX_N_SPILL_BYTES too small for CPUID boilerplate 1203195860 WARNING: unhandled syscall: unix:223 1204196528 need a error suppression for pthread_rwlock_init under os x? 1205197227 Support aio_* syscalls on Darwin 1206197456 valgrind should reject --suppressions=(directory) 1207197512 DWARF2 CFI reader: unhandled CFI instruction 0:10 1208197591 unhandled syscall 27 (mincore) 1209197793 Merge DCAS branch to the trunk == 85756, 142103 1210197794 Avoid duplicate filenames in Vex 1211197898 make check fails on current SVN 1212197901 make check fails also under exp-ptrcheck in current SVN 1213197929 Make --leak-resolution=high the default 1214197930 Reduce spacing between leak reports 1215197933 Print command line of client at start-up, and shorten preamble 1216197966 unhandled syscall 205 (x86-linux, --tool=exp-ptrcheck) 1217198395 add BBV to the distribution as an experimental tool 1218198624 Missing syscalls on Darwin: 82, 167, 281, 347 1219198649 callgrind_annotate doesn't cumulate counters 1220199338 callgrind_annotate sorting/thresholds are broken for all but Ir 1221199977 Valgrind complains about an unrecognized instruction in the 1222 atomic_incs test program 1223200029 valgrind isn't able to read Fedora 12 debuginfo 1224200760 darwin unhandled syscall: unix:284 1225200827 DRD doesn't work on Mac OS X 1226200990 VG_(read_millisecond_timer)() does not work correctly 1227201016 Valgrind does not support pthread_kill() on Mac OS 1228201169 Document --read-var-info 1229201323 Pre-3.5.0 performance sanity checking 1230201384 Review user manual for the 3.5.0 release 1231201585 mfpvr not implemented on ppc 1232201708 tests failing because x86 direction flag is left set 1233201757 Valgrind doesn't handle any recent sys_futex additions 1234204377 64-bit valgrind can not start a shell script 1235 (with #!/path/to/shell) if the shell is a 32-bit executable 1236n-i-bz drd: fixed assertion failure triggered by mutex reinitialization. 1237n-i-bz drd: fixed a bug that caused incorrect messages to be printed 1238 about memory allocation events with memory access tracing enabled 1239n-i-bz drd: fixed a memory leak triggered by vector clock deallocation 1240 1241(3.5.0: 19 Aug 2009, vex r1913, valgrind r10846). 1242 1243 1244 1245Release 3.4.1 (28 February 2009) 1246~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 12473.4.1 is a bug-fix release that fixes some regressions and assertion 1248failures in debug info reading in 3.4.0, most notably incorrect stack 1249traces on amd64-linux on older (glibc-2.3 based) systems. Various 1250other debug info problems are also fixed. A number of bugs in the 1251exp-ptrcheck tool introduced in 3.4.0 have been fixed. 1252 1253In view of the fact that 3.4.0 contains user-visible regressions 1254relative to 3.3.x, upgrading to 3.4.1 is recommended. Packagers are 1255encouraged to ship 3.4.1 in preference to 3.4.0. 1256 1257The fixed bugs are as follows. Note that "n-i-bz" stands for "not in 1258bugzilla" -- that is, a bug that was reported to us but never got a 1259bugzilla entry. We encourage you to file bugs in bugzilla 1260(http://bugs.kde.org/enter_valgrind_bug.cgi) rather than mailing the 1261developers (or mailing lists) directly -- bugs that are not entered 1262into bugzilla tend to get forgotten about or ignored. 1263 1264n-i-bz Fix various bugs reading icc-11 generated debug info 1265n-i-bz Fix various bugs reading gcc-4.4 generated debug info 1266n-i-bz Preliminary support for glibc-2.10 / Fedora 11 1267n-i-bz Cachegrind and Callgrind: handle non-power-of-two cache sizes, 1268 so as to support (eg) 24k Atom D1 and Core2 with 3/6/12MB L2. 1269179618 exp-ptrcheck crashed / exit prematurely 1270179624 helgrind: false positive races with pthread_create and 1271 recv/open/close/read 1272134207 pkg-config output contains @VG_PLATFORM@ 1273176926 floating point exception at valgrind startup with PPC 440EPX 1274181594 Bogus warning for empty text segment 1275173751 amd64->IR: 0x48 0xF 0x6F 0x45 (even more redundant rex prefixes) 1276181707 Dwarf3 doesn't require enumerations to have name 1277185038 exp-ptrcheck: "unhandled syscall: 285" (fallocate) on x86_64 1278185050 exp-ptrcheck: sg_main.c:727 (add_block_to_GlobalTree): 1279 Assertion '!already_present' failed. 1280185359 exp-ptrcheck unhandled syscall getresuid() 1281 1282(3.4.1.RC1: 24 Feb 2008, vex r1884, valgrind r9253). 1283(3.4.1: 28 Feb 2008, vex r1884, valgrind r9293). 1284 1285 1286 1287Release 3.4.0 (2 January 2009) 1288~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 12893.4.0 is a feature release with many significant improvements and the 1290usual collection of bug fixes. This release supports X86/Linux, 1291AMD64/Linux, PPC32/Linux and PPC64/Linux. Support for recent distros 1292(using gcc 4.4, glibc 2.8 and 2.9) has been added. 1293 12943.4.0 brings some significant tool improvements. Memcheck can now 1295report the origin of uninitialised values, the thread checkers 1296Helgrind and DRD are much improved, and we have a new experimental 1297tool, exp-Ptrcheck, which is able to detect overruns of stack and 1298global arrays. In detail: 1299 1300* Memcheck is now able to track the origin of uninitialised values. 1301 When it reports an uninitialised value error, it will try to show 1302 the origin of the value, as either a heap or stack allocation. 1303 Origin tracking is expensive and so is not enabled by default. To 1304 use it, specify --track-origins=yes. Memcheck's speed will be 1305 essentially halved, and memory usage will be significantly 1306 increased. Nevertheless it can drastically reduce the effort 1307 required to identify the root cause of uninitialised value errors, 1308 and so is often a programmer productivity win, despite running more 1309 slowly. 1310 1311* A version (1.4.0) of the Valkyrie GUI, that works with Memcheck in 1312 3.4.0, will be released shortly. 1313 1314* Helgrind's race detection algorithm has been completely redesigned 1315 and reimplemented, to address usability and scalability concerns: 1316 1317 - The new algorithm has a lower false-error rate: it is much less 1318 likely to report races that do not really exist. 1319 1320 - Helgrind will display full call stacks for both accesses involved 1321 in a race. This makes it easier to identify the root causes of 1322 races. 1323 1324 - Limitations on the size of program that can run have been removed. 1325 1326 - Performance has been modestly improved, although that is very 1327 workload-dependent. 1328 1329 - Direct support for Qt4 threading has been added. 1330 1331 - pthread_barriers are now directly supported. 1332 1333 - Helgrind works well on all supported Linux targets. 1334 1335* The DRD thread debugging tool has seen major improvements: 1336 1337 - Greatly improved performance and significantly reduced memory 1338 usage. 1339 1340 - Support for several major threading libraries (Boost.Thread, Qt4, 1341 glib, OpenMP) has been added. 1342 1343 - Support for atomic instructions, POSIX semaphores, barriers and 1344 reader-writer locks has been added. 1345 1346 - Works now on PowerPC CPUs too. 1347 1348 - Added support for printing thread stack usage at thread exit time. 1349 1350 - Added support for debugging lock contention. 1351 1352 - Added a manual for Drd. 1353 1354* A new experimental tool, exp-Ptrcheck, has been added. Ptrcheck 1355 checks for misuses of pointers. In that sense it is a bit like 1356 Memcheck. However, Ptrcheck can do things Memcheck can't: it can 1357 detect overruns of stack and global arrays, it can detect 1358 arbitrarily far out-of-bounds accesses to heap blocks, and it can 1359 detect accesses heap blocks that have been freed a very long time 1360 ago (millions of blocks in the past). 1361 1362 Ptrcheck currently works only on x86-linux and amd64-linux. To use 1363 it, use --tool=exp-ptrcheck. A simple manual is provided, as part 1364 of the main Valgrind documentation. As this is an experimental 1365 tool, we would be particularly interested in hearing about your 1366 experiences with it. 1367 1368* exp-Omega, an experimental instantaneous leak-detecting tool, is no 1369 longer built by default, although the code remains in the repository 1370 and the tarball. This is due to three factors: a perceived lack of 1371 users, a lack of maintenance, and concerns that it may not be 1372 possible to achieve reliable operation using the existing design. 1373 1374* As usual, support for the latest Linux distros and toolchain 1375 components has been added. It should work well on Fedora Core 10, 1376 OpenSUSE 11.1 and Ubuntu 8.10. gcc-4.4 (in its current pre-release 1377 state) is supported, as is glibc-2.9. The C++ demangler has been 1378 updated so as to work well with C++ compiled by even the most recent 1379 g++'s. 1380 1381* You can now use frame-level wildcards in suppressions. This was a 1382 frequently-requested enhancement. A line "..." in a suppression now 1383 matches zero or more frames. This makes it easier to write 1384 suppressions which are precise yet insensitive to changes in 1385 inlining behaviour. 1386 1387* 3.4.0 adds support on x86/amd64 for the SSSE3 instruction set. 1388 1389* Very basic support for IBM Power6 has been added (64-bit processes only). 1390 1391* Valgrind is now cross-compilable. For example, it is possible to 1392 cross compile Valgrind on an x86/amd64-linux host, so that it runs 1393 on a ppc32/64-linux target. 1394 1395* You can set the main thread's stack size at startup using the 1396 new --main-stacksize= flag (subject of course to ulimit settings). 1397 This is useful for running apps that need a lot of stack space. 1398 1399* The limitation that you can't use --trace-children=yes together 1400 with --db-attach=yes has been removed. 1401 1402* The following bugs have been fixed. Note that "n-i-bz" stands for 1403 "not in bugzilla" -- that is, a bug that was reported to us but 1404 never got a bugzilla entry. We encourage you to file bugs in 1405 bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 1406 mailing the developers (or mailing lists) directly. 1407 1408 n-i-bz Make return types for some client requests 64-bit clean 1409 n-i-bz glibc 2.9 support 1410 n-i-bz ignore unsafe .valgrindrc's (CVE-2008-4865) 1411 n-i-bz MPI_Init(0,0) is valid but libmpiwrap.c segfaults 1412 n-i-bz Building in an env without gdb gives bogus gdb attach 1413 92456 Tracing the origin of uninitialised memory 1414 106497 Valgrind does not demangle some C++ template symbols 1415 162222 ==106497 1416 151612 Suppression with "..." (frame-level wildcards in .supp files) 1417 156404 Unable to start oocalc under memcheck on openSUSE 10.3 (64-bit) 1418 159285 unhandled syscall:25 (stime, on x86-linux) 1419 159452 unhandled ioctl 0x8B01 on "valgrind iwconfig" 1420 160954 ppc build of valgrind crashes with illegal instruction (isel) 1421 160956 mallinfo implementation, w/ patch 1422 162092 Valgrind fails to start gnome-system-monitor 1423 162819 malloc_free_fill test doesn't pass on glibc2.8 x86 1424 163794 assertion failure with "--track-origins=yes" 1425 163933 sigcontext.err and .trapno must be set together 1426 163955 remove constraint !(--db-attach=yes && --trace-children=yes) 1427 164476 Missing kernel module loading system calls 1428 164669 SVN regression: mmap() drops posix file locks 1429 166581 Callgrind output corruption when program forks 1430 167288 Patch file for missing system calls on Cell BE 1431 168943 unsupported scas instruction pentium 1432 171645 Unrecognised instruction (MOVSD, non-binutils encoding) 1433 172417 x86->IR: 0x82 ... 1434 172563 amd64->IR: 0xD9 0xF5 - fprem1 1435 173099 .lds linker script generation error 1436 173177 [x86_64] syscalls: 125/126/179 (capget/capset/quotactl) 1437 173751 amd64->IR: 0x48 0xF 0x6F 0x45 (even more redundant prefixes) 1438 174532 == 173751 1439 174908 --log-file value not expanded correctly for core file 1440 175044 Add lookup_dcookie for amd64 1441 175150 x86->IR: 0xF2 0xF 0x11 0xC1 (movss non-binutils encoding) 1442 1443Developer-visible changes: 1444 1445* Valgrind's debug-info reading machinery has been majorly overhauled. 1446 It can now correctly establish the addresses for ELF data symbols, 1447 which is something that has never worked properly before now. 1448 1449 Also, Valgrind can now read DWARF3 type and location information for 1450 stack and global variables. This makes it possible to use the 1451 framework to build tools that rely on knowing the type and locations 1452 of stack and global variables, for example exp-Ptrcheck. 1453 1454 Reading of such information is disabled by default, because most 1455 tools don't need it, and because it is expensive in space and time. 1456 However, you can force Valgrind to read it, using the 1457 --read-var-info=yes flag. Memcheck, Helgrind and DRD are able to 1458 make use of such information, if present, to provide source-level 1459 descriptions of data addresses in the error messages they create. 1460 1461(3.4.0.RC1: 24 Dec 2008, vex r1878, valgrind r8882). 1462(3.4.0: 3 Jan 2009, vex r1878, valgrind r8899). 1463 1464 1465 1466Release 3.3.1 (4 June 2008) 1467~~~~~~~~~~~~~~~~~~~~~~~~~~~ 14683.3.1 fixes a bunch of bugs in 3.3.0, adds support for glibc-2.8 based 1469systems (openSUSE 11, Fedora Core 9), improves the existing glibc-2.7 1470support, and adds support for the SSSE3 (Core 2) instruction set. 1471 14723.3.1 will likely be the last release that supports some very old 1473systems. In particular, the next major release, 3.4.0, will drop 1474support for the old LinuxThreads threading library, and for gcc 1475versions prior to 3.0. 1476 1477The fixed bugs are as follows. Note that "n-i-bz" stands for "not in 1478bugzilla" -- that is, a bug that was reported to us but never got a 1479bugzilla entry. We encourage you to file bugs in bugzilla 1480(http://bugs.kde.org/enter_valgrind_bug.cgi) rather than mailing the 1481developers (or mailing lists) directly -- bugs that are not entered 1482into bugzilla tend to get forgotten about or ignored. 1483 1484n-i-bz Massif segfaults at exit 1485n-i-bz Memcheck asserts on Altivec code 1486n-i-bz fix sizeof bug in Helgrind 1487n-i-bz check fd on sys_llseek 1488n-i-bz update syscall lists to kernel 2.6.23.1 1489n-i-bz support sys_sync_file_range 1490n-i-bz handle sys_sysinfo, sys_getresuid, sys_getresgid on ppc64-linux 1491n-i-bz intercept memcpy in 64-bit ld.so's 1492n-i-bz Fix wrappers for sys_{futimesat,utimensat} 1493n-i-bz Minor false-error avoidance fixes for Memcheck 1494n-i-bz libmpiwrap.c: add a wrapper for MPI_Waitany 1495n-i-bz helgrind support for glibc-2.8 1496n-i-bz partial fix for mc_leakcheck.c:698 assert: 1497 'lc_shadows[i]->data + lc_shadows[i] ... 1498n-i-bz Massif/Cachegrind output corruption when programs fork 1499n-i-bz register allocator fix: handle spill stores correctly 1500n-i-bz add support for PA6T PowerPC CPUs 1501126389 vex x86->IR: 0xF 0xAE (FXRSTOR) 1502158525 ==126389 1503152818 vex x86->IR: 0xF3 0xAC (repz lodsb) 1504153196 vex x86->IR: 0xF2 0xA6 (repnz cmpsb) 1505155011 vex x86->IR: 0xCF (iret) 1506155091 Warning [...] unhandled DW_OP_ opcode 0x23 1507156960 ==155901 1508155528 support Core2/SSSE3 insns on x86/amd64 1509155929 ms_print fails on massif outputs containing long lines 1510157665 valgrind fails on shmdt(0) after shmat to 0 1511157748 support x86 PUSHFW/POPFW 1512158212 helgrind: handle pthread_rwlock_try{rd,wr}lock. 1513158425 sys_poll incorrectly emulated when RES==0 1514158744 vex amd64->IR: 0xF0 0x41 0xF 0xC0 (xaddb) 1515160907 Support for a couple of recent Linux syscalls 1516161285 Patch -- support for eventfd() syscall 1517161378 illegal opcode in debug libm (FUCOMPP) 1518160136 ==161378 1519161487 number of suppressions files is limited to 10 1520162386 ms_print typo in milliseconds time unit for massif 1521161036 exp-drd: client allocated memory was never freed 1522162663 signalfd_wrapper fails on 64bit linux 1523 1524(3.3.1.RC1: 2 June 2008, vex r1854, valgrind r8169). 1525(3.3.1: 4 June 2008, vex r1854, valgrind r8180). 1526 1527 1528 1529Release 3.3.0 (7 December 2007) 1530~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 15313.3.0 is a feature release with many significant improvements and the 1532usual collection of bug fixes. This release supports X86/Linux, 1533AMD64/Linux, PPC32/Linux and PPC64/Linux. Support for recent distros 1534(using gcc 4.3, glibc 2.6 and 2.7) has been added. 1535 1536The main excitement in 3.3.0 is new and improved tools. Helgrind 1537works again, Massif has been completely overhauled and much improved, 1538Cachegrind now does branch-misprediction profiling, and a new category 1539of experimental tools has been created, containing two new tools: 1540Omega and DRD. There are many other smaller improvements. In detail: 1541 1542- Helgrind has been completely overhauled and works for the first time 1543 since Valgrind 2.2.0. Supported functionality is: detection of 1544 misuses of the POSIX PThreads API, detection of potential deadlocks 1545 resulting from cyclic lock dependencies, and detection of data 1546 races. Compared to the 2.2.0 Helgrind, the race detection algorithm 1547 has some significant improvements aimed at reducing the false error 1548 rate. Handling of various kinds of corner cases has been improved. 1549 Efforts have been made to make the error messages easier to 1550 understand. Extensive documentation is provided. 1551 1552- Massif has been completely overhauled. Instead of measuring 1553 space-time usage -- which wasn't always useful and many people found 1554 confusing -- it now measures space usage at various points in the 1555 execution, including the point of peak memory allocation. Its 1556 output format has also changed: instead of producing PostScript 1557 graphs and HTML text, it produces a single text output (via the new 1558 'ms_print' script) that contains both a graph and the old textual 1559 information, but in a more compact and readable form. Finally, the 1560 new version should be more reliable than the old one, as it has been 1561 tested more thoroughly. 1562 1563- Cachegrind has been extended to do branch-misprediction profiling. 1564 Both conditional and indirect branches are profiled. The default 1565 behaviour of Cachegrind is unchanged. To use the new functionality, 1566 give the option --branch-sim=yes. 1567 1568- A new category of "experimental tools" has been created. Such tools 1569 may not work as well as the standard tools, but are included because 1570 some people will find them useful, and because exposure to a wider 1571 user group provides tool authors with more end-user feedback. These 1572 tools have a "exp-" prefix attached to their names to indicate their 1573 experimental nature. Currently there are two experimental tools: 1574 1575 * exp-Omega: an instantaneous leak detector. See 1576 exp-omega/docs/omega_introduction.txt. 1577 1578 * exp-DRD: a data race detector based on the happens-before 1579 relation. See exp-drd/docs/README.txt. 1580 1581- Scalability improvements for very large programs, particularly those 1582 which have a million or more malloc'd blocks in use at once. These 1583 improvements mostly affect Memcheck. Memcheck is also up to 10% 1584 faster for all programs, with x86-linux seeing the largest 1585 improvement. 1586 1587- Works well on the latest Linux distros. Has been tested on Fedora 1588 Core 8 (x86, amd64, ppc32, ppc64) and openSUSE 10.3. glibc 2.6 and 1589 2.7 are supported. gcc-4.3 (in its current pre-release state) is 1590 supported. At the same time, 3.3.0 retains support for older 1591 distros. 1592 1593- The documentation has been modestly reorganised with the aim of 1594 making it easier to find information on common-usage scenarios. 1595 Some advanced material has been moved into a new chapter in the main 1596 manual, so as to unclutter the main flow, and other tidying up has 1597 been done. 1598 1599- There is experimental support for AIX 5.3, both 32-bit and 64-bit 1600 processes. You need to be running a 64-bit kernel to use Valgrind 1601 on a 64-bit executable. 1602 1603- There have been some changes to command line options, which may 1604 affect you: 1605 1606 * --log-file-exactly and 1607 --log-file-qualifier options have been removed. 1608 1609 To make up for this --log-file option has been made more powerful. 1610 It now accepts a %p format specifier, which is replaced with the 1611 process ID, and a %q{FOO} format specifier, which is replaced with 1612 the contents of the environment variable FOO. 1613 1614 * --child-silent-after-fork=yes|no [no] 1615 1616 Causes Valgrind to not show any debugging or logging output for 1617 the child process resulting from a fork() call. This can make the 1618 output less confusing (although more misleading) when dealing with 1619 processes that create children. 1620 1621 * --cachegrind-out-file, --callgrind-out-file and --massif-out-file 1622 1623 These control the names of the output files produced by 1624 Cachegrind, Callgrind and Massif. They accept the same %p and %q 1625 format specifiers that --log-file accepts. --callgrind-out-file 1626 replaces Callgrind's old --base option. 1627 1628 * Cachegrind's 'cg_annotate' script no longer uses the --<pid> 1629 option to specify the output file. Instead, the first non-option 1630 argument is taken to be the name of the output file, and any 1631 subsequent non-option arguments are taken to be the names of 1632 source files to be annotated. 1633 1634 * Cachegrind and Callgrind now use directory names where possible in 1635 their output files. This means that the -I option to 1636 'cg_annotate' and 'callgrind_annotate' should not be needed in 1637 most cases. It also means they can correctly handle the case 1638 where two source files in different directories have the same 1639 name. 1640 1641- Memcheck offers a new suppression kind: "Jump". This is for 1642 suppressing jump-to-invalid-address errors. Previously you had to 1643 use an "Addr1" suppression, which didn't make much sense. 1644 1645- Memcheck has new flags --malloc-fill=<hexnum> and 1646 --free-fill=<hexnum> which free malloc'd / free'd areas with the 1647 specified byte. This can help shake out obscure memory corruption 1648 problems. The definedness and addressability of these areas is 1649 unchanged -- only the contents are affected. 1650 1651- The behaviour of Memcheck's client requests VALGRIND_GET_VBITS and 1652 VALGRIND_SET_VBITS have changed slightly. They no longer issue 1653 addressability errors -- if either array is partially unaddressable, 1654 they just return 3 (as before). Also, SET_VBITS doesn't report 1655 definedness errors if any of the V bits are undefined. 1656 1657- The following Memcheck client requests have been removed: 1658 VALGRIND_MAKE_NOACCESS 1659 VALGRIND_MAKE_WRITABLE 1660 VALGRIND_MAKE_READABLE 1661 VALGRIND_CHECK_WRITABLE 1662 VALGRIND_CHECK_READABLE 1663 VALGRIND_CHECK_DEFINED 1664 They were deprecated in 3.2.0, when equivalent but better-named client 1665 requests were added. See the 3.2.0 release notes for more details. 1666 1667- The behaviour of the tool Lackey has changed slightly. First, the output 1668 from --trace-mem has been made more compact, to reduce the size of the 1669 traces. Second, a new option --trace-superblocks has been added, which 1670 shows the addresses of superblocks (code blocks) as they are executed. 1671 1672- The following bugs have been fixed. Note that "n-i-bz" stands for 1673 "not in bugzilla" -- that is, a bug that was reported to us but 1674 never got a bugzilla entry. We encourage you to file bugs in 1675 bugzilla (http://bugs.kde.org/enter_valgrind_bug.cgi) rather than 1676 mailing the developers (or mailing lists) directly. 1677 1678 n-i-bz x86_linux_REDIR_FOR_index() broken 1679 n-i-bz guest-amd64/toIR.c:2512 (dis_op2_E_G): Assertion `0' failed. 1680 n-i-bz Support x86 INT insn (INT (0xCD) 0x40 - 0x43) 1681 n-i-bz Add sys_utimensat system call for Linux x86 platform 1682 79844 Helgrind complains about race condition which does not exist 1683 82871 Massif output function names too short 1684 89061 Massif: ms_main.c:485 (get_XCon): Assertion `xpt->max_chi...' 1685 92615 Write output from Massif at crash 1686 95483 massif feature request: include peak allocation in report 1687 112163 MASSIF crashed with signal 7 (SIGBUS) after running 2 days 1688 119404 problems running setuid executables (partial fix) 1689 121629 add instruction-counting mode for timing 1690 127371 java vm giving unhandled instruction bytes: 0x26 0x2E 0x64 0x65 1691 129937 ==150380 1692 129576 Massif loses track of memory, incorrect graphs 1693 132132 massif --format=html output does not do html entity escaping 1694 132950 Heap alloc/usage summary 1695 133962 unhandled instruction bytes: 0xF2 0x4C 0xF 0x10 1696 134990 use -fno-stack-protector if possible 1697 136382 ==134990 1698 137396 I would really like helgrind to work again... 1699 137714 x86/amd64->IR: 0x66 0xF 0xF7 0xC6 (maskmovq, maskmovdq) 1700 141631 Massif: percentages don't add up correctly 1701 142706 massif numbers don't seem to add up 1702 143062 massif crashes on app exit with signal 8 SIGFPE 1703 144453 (get_XCon): Assertion 'xpt->max_children != 0' failed. 1704 145559 valgrind aborts when malloc_stats is called 1705 145609 valgrind aborts all runs with 'repeated section!' 1706 145622 --db-attach broken again on x86-64 1707 145837 ==149519 1708 145887 PPC32: getitimer() system call is not supported 1709 146252 ==150678 1710 146456 (update_XCon): Assertion 'xpt->curr_space >= -space_delta'... 1711 146701 ==134990 1712 146781 Adding support for private futexes 1713 147325 valgrind internal error on syscall (SYS_io_destroy, 0) 1714 147498 amd64->IR: 0xF0 0xF 0xB0 0xF (lock cmpxchg %cl,(%rdi)) 1715 147545 Memcheck: mc_main.c:817 (get_sec_vbits8): Assertion 'n' failed. 1716 147628 SALC opcode 0xd6 unimplemented 1717 147825 crash on amd64-linux with gcc 4.2 and glibc 2.6 (CFI) 1718 148174 Incorrect type of freed_list_volume causes assertion [...] 1719 148447 x86_64 : new NOP codes: 66 66 66 66 2e 0f 1f 1720 149182 PPC Trap instructions not implemented in valgrind 1721 149504 Assertion hit on alloc_xpt->curr_space >= -space_delta 1722 149519 ppc32: V aborts with SIGSEGV on execution of a signal handler 1723 149892 ==137714 1724 150044 SEGV during stack deregister 1725 150380 dwarf/gcc interoperation (dwarf3 read problems) 1726 150408 ==148447 1727 150678 guest-amd64/toIR.c:3741 (dis_Grp5): Assertion `sz == 4' failed 1728 151209 V unable to execute programs for users with UID > 2^16 1729 151938 help on --db-command= misleading 1730 152022 subw $0x28, %%sp causes assertion failure in memcheck 1731 152357 inb and outb not recognized in 64-bit mode 1732 152501 vex x86->IR: 0x27 0x66 0x89 0x45 (daa) 1733 152818 vex x86->IR: 0xF3 0xAC 0xFC 0x9C (rep lodsb) 1734 1735Developer-visible changes: 1736 1737- The names of some functions and types within the Vex IR have 1738 changed. Run 'svn log -r1689 VEX/pub/libvex_ir.h' for full details. 1739 Any existing standalone tools will have to be updated to reflect 1740 these changes. The new names should be clearer. The file 1741 VEX/pub/libvex_ir.h is also much better commented. 1742 1743- A number of new debugging command line options have been added. 1744 These are mostly of use for debugging the symbol table and line 1745 number readers: 1746 1747 --trace-symtab-patt=<patt> limit debuginfo tracing to obj name <patt> 1748 --trace-cfi=no|yes show call-frame-info details? [no] 1749 --debug-dump=syms mimic /usr/bin/readelf --syms 1750 --debug-dump=line mimic /usr/bin/readelf --debug-dump=line 1751 --debug-dump=frames mimic /usr/bin/readelf --debug-dump=frames 1752 --sym-offsets=yes|no show syms in form 'name+offset' ? [no] 1753 1754- Internally, the code base has been further factorised and 1755 abstractified, particularly with respect to support for non-Linux 1756 OSs. 1757 1758(3.3.0.RC1: 2 Dec 2007, vex r1803, valgrind r7268). 1759(3.3.0.RC2: 5 Dec 2007, vex r1804, valgrind r7282). 1760(3.3.0.RC3: 9 Dec 2007, vex r1804, valgrind r7288). 1761(3.3.0: 10 Dec 2007, vex r1804, valgrind r7290). 1762 1763 1764 1765Release 3.2.3 (29 Jan 2007) 1766~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1767Unfortunately 3.2.2 introduced a regression which can cause an 1768assertion failure ("vex: the `impossible' happened: eqIRConst") when 1769running obscure pieces of SSE code. 3.2.3 fixes this and adds one 1770more glibc-2.5 intercept. In all other respects it is identical to 17713.2.2. Please do not use (or package) 3.2.2; instead use 3.2.3. 1772 1773n-i-bz vex: the `impossible' happened: eqIRConst 1774n-i-bz Add an intercept for glibc-2.5 __stpcpy_chk 1775 1776(3.2.3: 29 Jan 2007, vex r1732, valgrind r6560). 1777 1778 1779Release 3.2.2 (22 Jan 2007) 1780~~~~~~~~~~~~~~~~~~~~~~~~~~~ 17813.2.2 fixes a bunch of bugs in 3.2.1, adds support for glibc-2.5 based 1782systems (openSUSE 10.2, Fedora Core 6), improves support for icc-9.X 1783compiled code, and brings modest performance improvements in some 1784areas, including amd64 floating point, powerpc support, and startup 1785responsiveness on all targets. 1786 1787The fixed bugs are as follows. Note that "n-i-bz" stands for "not in 1788bugzilla" -- that is, a bug that was reported to us but never got a 1789bugzilla entry. We encourage you to file bugs in bugzilla 1790(http://bugs.kde.org/enter_valgrind_bug.cgi) rather than mailing the 1791developers (or mailing lists) directly. 1792 1793129390 ppc?->IR: some kind of VMX prefetch (dstt) 1794129968 amd64->IR: 0xF 0xAE 0x0 (fxsave) 1795134319 ==129968 1796133054 'make install' fails with syntax errors 1797118903 ==133054 1798132998 startup fails in when running on UML 1799134207 pkg-config output contains @VG_PLATFORM@ 1800134727 valgrind exits with "Value too large for defined data type" 1801n-i-bz ppc32/64: support mcrfs 1802n-i-bz Cachegrind/Callgrind: Update cache parameter detection 1803135012 x86->IR: 0xD7 0x8A 0xE0 0xD0 (xlat) 1804125959 ==135012 1805126147 x86->IR: 0xF2 0xA5 0xF 0x77 (repne movsw) 1806136650 amd64->IR: 0xC2 0x8 0x0 1807135421 x86->IR: unhandled Grp5(R) case 6 1808n-i-bz Improved documentation of the IR intermediate representation 1809n-i-bz jcxz (x86) (users list, 8 Nov) 1810n-i-bz ExeContext hashing fix 1811n-i-bz fix CFI reading failures ("Dwarf CFI 0:24 0:32 0:48 0:7") 1812n-i-bz fix Cachegrind/Callgrind simulation bug 1813n-i-bz libmpiwrap.c: fix handling of MPI_LONG_DOUBLE 1814n-i-bz make User errors suppressible 1815136844 corrupted malloc line when using --gen-suppressions=yes 1816138507 ==136844 1817n-i-bz Speed up the JIT's register allocator 1818n-i-bz Fix confusing leak-checker flag hints 1819n-i-bz Support recent autoswamp versions 1820n-i-bz ppc32/64 dispatcher speedups 1821n-i-bz ppc64 front end rld/rlw improvements 1822n-i-bz ppc64 back end imm64 improvements 1823136300 support 64K pages on ppc64-linux 1824139124 == 136300 1825n-i-bz fix ppc insn set tests for gcc >= 4.1 1826137493 x86->IR: recent binutils no-ops 1827137714 x86->IR: 0x66 0xF 0xF7 0xC6 (maskmovdqu) 1828138424 "failed in UME with error 22" (produce a better error msg) 1829138856 ==138424 1830138627 Enhancement support for prctl ioctls 1831138896 Add support for usb ioctls 1832136059 ==138896 1833139050 ppc32->IR: mfspr 268/269 instructions not handled 1834n-i-bz ppc32->IR: lvxl/stvxl 1835n-i-bz glibc-2.5 support 1836n-i-bz memcheck: provide replacement for mempcpy 1837n-i-bz memcheck: replace bcmp in ld.so 1838n-i-bz Use 'ifndef' in VEX's Makefile correctly 1839n-i-bz Suppressions for MVL 4.0.1 on ppc32-linux 1840n-i-bz libmpiwrap.c: Fixes for MPICH 1841n-i-bz More robust handling of hinted client mmaps 1842139776 Invalid read in unaligned memcpy with Intel compiler v9 1843n-i-bz Generate valid XML even for very long fn names 1844n-i-bz Don't prompt about suppressions for unshown reachable leaks 1845139910 amd64 rcl is not supported 1846n-i-bz DWARF CFI reader: handle DW_CFA_undefined 1847n-i-bz DWARF CFI reader: handle icc9 generated CFI info better 1848n-i-bz fix false uninit-value errs in icc9 generated FP code 1849n-i-bz reduce extraneous frames in libmpiwrap.c 1850n-i-bz support pselect6 on amd64-linux 1851 1852(3.2.2: 22 Jan 2007, vex r1729, valgrind r6545). 1853 1854 1855Release 3.2.1 (16 Sept 2006) 1856~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 18573.2.1 adds x86/amd64 support for all SSE3 instructions except monitor 1858and mwait, further reduces memcheck's false error rate on all 1859platforms, adds support for recent binutils (in OpenSUSE 10.2 and 1860Fedora Rawhide) and fixes a bunch of bugs in 3.2.0. Some of the fixed 1861bugs were causing large programs to segfault with --tool=callgrind and 1862--tool=cachegrind, so an upgrade is recommended. 1863 1864In view of the fact that any 3.3.0 release is unlikely to happen until 1865well into 1Q07, we intend to keep the 3.2.X line alive for a while 1866yet, and so we tentatively plan a 3.2.2 release sometime in December 186706. 1868 1869The fixed bugs are as follows. Note that "n-i-bz" stands for "not in 1870bugzilla" -- that is, a bug that was reported to us but never got a 1871bugzilla entry. 1872 1873n-i-bz Expanding brk() into last available page asserts 1874n-i-bz ppc64-linux stack RZ fast-case snafu 1875n-i-bz 'c' in --gen-supps=yes doesn't work 1876n-i-bz VG_N_SEGMENTS too low (users, 28 June) 1877n-i-bz VG_N_SEGNAMES too low (Stu Robinson) 1878106852 x86->IR: fisttp (SSE3) 1879117172 FUTEX_WAKE does not use uaddr2 1880124039 Lacks support for VKI_[GP]IO_UNIMAP* 1881127521 amd64->IR: 0xF0 0x48 0xF 0xC7 (cmpxchg8b) 1882128917 amd64->IR: 0x66 0xF 0xF6 0xC4 (psadbw,SSE2) 1883129246 JJ: ppc32/ppc64 syscalls, w/ patch 1884129358 x86->IR: fisttpl (SSE3) 1885129866 cachegrind/callgrind causes executable to die 1886130020 Can't stat .so/.exe error while reading symbols 1887130388 Valgrind aborts when process calls malloc_trim() 1888130638 PATCH: ppc32 missing system calls 1889130785 amd64->IR: unhandled instruction "pushfq" 1890131481: (HINT_NOP) vex x86->IR: 0xF 0x1F 0x0 0xF 1891131298 ==131481 1892132146 Programs with long sequences of bswap[l,q]s 1893132918 vex amd64->IR: 0xD9 0xF8 (fprem) 1894132813 Assertion at priv/guest-x86/toIR.c:652 fails 1895133051 'cfsi->len > 0 && cfsi->len < 2000000' failed 1896132722 valgrind header files are not standard C 1897n-i-bz Livelocks entire machine (users list, Timothy Terriberry) 1898n-i-bz Alex Bennee mmap problem (9 Aug) 1899n-i-bz BartV: Don't print more lines of a stack-trace than were obtained. 1900n-i-bz ppc32 SuSE 10.1 redir 1901n-i-bz amd64 padding suppressions 1902n-i-bz amd64 insn printing fix. 1903n-i-bz ppc cmp reg,reg fix 1904n-i-bz x86/amd64 iropt e/rflag reduction rules 1905n-i-bz SuSE 10.1 (ppc32) minor fixes 1906133678 amd64->IR: 0x48 0xF 0xC5 0xC0 (pextrw?) 1907133694 aspacem assertion: aspacem_minAddr <= holeStart 1908n-i-bz callgrind: fix warning about malformed creator line 1909n-i-bz callgrind: fix annotate script for data produced with 1910 --dump-instr=yes 1911n-i-bz callgrind: fix failed assertion when toggling 1912 instrumentation mode 1913n-i-bz callgrind: fix annotate script fix warnings with 1914 --collect-jumps=yes 1915n-i-bz docs path hardwired (Dennis Lubert) 1916 1917The following bugs were not fixed, due primarily to lack of developer 1918time, and also because bug reporters did not answer requests for 1919feedback in time for the release: 1920 1921129390 ppc?->IR: some kind of VMX prefetch (dstt) 1922129968 amd64->IR: 0xF 0xAE 0x0 (fxsave) 1923133054 'make install' fails with syntax errors 1924n-i-bz Signal race condition (users list, 13 June, Johannes Berg) 1925n-i-bz Unrecognised instruction at address 0x70198EC2 (users list, 1926 19 July, Bennee) 1927132998 startup fails in when running on UML 1928 1929The following bug was tentatively fixed on the mainline but the fix 1930was considered too risky to push into 3.2.X: 1931 1932133154 crash when using client requests to register/deregister stack 1933 1934(3.2.1: 16 Sept 2006, vex r1658, valgrind r6070). 1935 1936 1937Release 3.2.0 (7 June 2006) 1938~~~~~~~~~~~~~~~~~~~~~~~~~~~ 19393.2.0 is a feature release with many significant improvements and the 1940usual collection of bug fixes. This release supports X86/Linux, 1941AMD64/Linux, PPC32/Linux and PPC64/Linux. 1942 1943Performance, especially of Memcheck, is improved, Addrcheck has been 1944removed, Callgrind has been added, PPC64/Linux support has been added, 1945Lackey has been improved, and MPI support has been added. In detail: 1946 1947- Memcheck has improved speed and reduced memory use. Run times are 1948 typically reduced by 15-30%, averaging about 24% for SPEC CPU2000. 1949 The other tools have smaller but noticeable speed improvments. We 1950 are interested to hear what improvements users get. 1951 1952 Memcheck uses less memory due to the introduction of a compressed 1953 representation for shadow memory. The space overhead has been 1954 reduced by a factor of up to four, depending on program behaviour. 1955 This means you should be able to run programs that use more memory 1956 than before without hitting problems. 1957 1958- Addrcheck has been removed. It has not worked since version 2.4.0, 1959 and the speed and memory improvements to Memcheck make it redundant. 1960 If you liked using Addrcheck because it didn't give undefined value 1961 errors, you can use the new Memcheck option --undef-value-errors=no 1962 to get the same behaviour. 1963 1964- The number of undefined-value errors incorrectly reported by 1965 Memcheck has been reduced (such false reports were already very 1966 rare). In particular, efforts have been made to ensure Memcheck 1967 works really well with gcc 4.0/4.1-generated code on X86/Linux and 1968 AMD64/Linux. 1969 1970- Josef Weidendorfer's popular Callgrind tool has been added. Folding 1971 it in was a logical step given its popularity and usefulness, and 1972 makes it easier for us to ensure it works "out of the box" on all 1973 supported targets. The associated KDE KCachegrind GUI remains a 1974 separate project. 1975 1976- A new release of the Valkyrie GUI for Memcheck, version 1.2.0, 1977 accompanies this release. Improvements over previous releases 1978 include improved robustness, many refinements to the user interface, 1979 and use of a standard autoconf/automake build system. You can get 1980 it from http://www.valgrind.org/downloads/guis.html. 1981 1982- Valgrind now works on PPC64/Linux. As with the AMD64/Linux port, 1983 this supports programs using to 32G of address space. On 64-bit 1984 capable PPC64/Linux setups, you get a dual architecture build so 1985 that both 32-bit and 64-bit executables can be run. Linux on POWER5 1986 is supported, and POWER4 is also believed to work. Both 32-bit and 1987 64-bit DWARF2 is supported. This port is known to work well with 1988 both gcc-compiled and xlc/xlf-compiled code. 1989 1990- Floating point accuracy has been improved for PPC32/Linux. 1991 Specifically, the floating point rounding mode is observed on all FP 1992 arithmetic operations, and multiply-accumulate instructions are 1993 preserved by the compilation pipeline. This means you should get FP 1994 results which are bit-for-bit identical to a native run. These 1995 improvements are also present in the PPC64/Linux port. 1996 1997- Lackey, the example tool, has been improved: 1998 1999 * It has a new option --detailed-counts (off by default) which 2000 causes it to print out a count of loads, stores and ALU operations 2001 done, and their sizes. 2002 2003 * It has a new option --trace-mem (off by default) which causes it 2004 to print out a trace of all memory accesses performed by a 2005 program. It's a good starting point for building Valgrind tools 2006 that need to track memory accesses. Read the comments at the top 2007 of the file lackey/lk_main.c for details. 2008 2009 * The original instrumentation (counting numbers of instructions, 2010 jumps, etc) is now controlled by a new option --basic-counts. It 2011 is on by default. 2012 2013- MPI support: partial support for debugging distributed applications 2014 using the MPI library specification has been added. Valgrind is 2015 aware of the memory state changes caused by a subset of the MPI 2016 functions, and will carefully check data passed to the (P)MPI_ 2017 interface. 2018 2019- A new flag, --error-exitcode=, has been added. This allows changing 2020 the exit code in runs where Valgrind reported errors, which is 2021 useful when using Valgrind as part of an automated test suite. 2022 2023- Various segfaults when reading old-style "stabs" debug information 2024 have been fixed. 2025 2026- A simple performance evaluation suite has been added. See 2027 perf/README and README_DEVELOPERS for details. There are 2028 various bells and whistles. 2029 2030- New configuration flags: 2031 --enable-only32bit 2032 --enable-only64bit 2033 By default, on 64 bit platforms (ppc64-linux, amd64-linux) the build 2034 system will attempt to build a Valgrind which supports both 32-bit 2035 and 64-bit executables. This may not be what you want, and you can 2036 override the default behaviour using these flags. 2037 2038Please note that Helgrind is still not working. We have made an 2039important step towards making it work again, however, with the 2040addition of function wrapping (see below). 2041 2042Other user-visible changes: 2043 2044- Valgrind now has the ability to intercept and wrap arbitrary 2045 functions. This is a preliminary step towards making Helgrind work 2046 again, and was required for MPI support. 2047 2048- There are some changes to Memcheck's client requests. Some of them 2049 have changed names: 2050 2051 MAKE_NOACCESS --> MAKE_MEM_NOACCESS 2052 MAKE_WRITABLE --> MAKE_MEM_UNDEFINED 2053 MAKE_READABLE --> MAKE_MEM_DEFINED 2054 2055 CHECK_WRITABLE --> CHECK_MEM_IS_ADDRESSABLE 2056 CHECK_READABLE --> CHECK_MEM_IS_DEFINED 2057 CHECK_DEFINED --> CHECK_VALUE_IS_DEFINED 2058 2059 The reason for the change is that the old names are subtly 2060 misleading. The old names will still work, but they are deprecated 2061 and may be removed in a future release. 2062 2063 We also added a new client request: 2064 2065 MAKE_MEM_DEFINED_IF_ADDRESSABLE(a, len) 2066 2067 which is like MAKE_MEM_DEFINED but only affects a byte if the byte is 2068 already addressable. 2069 2070- The way client requests are encoded in the instruction stream has 2071 changed. Unfortunately, this means 3.2.0 will not honour client 2072 requests compiled into binaries using headers from earlier versions 2073 of Valgrind. We will try to keep the client request encodings more 2074 stable in future. 2075 2076BUGS FIXED: 2077 2078108258 NPTL pthread cleanup handlers not called 2079117290 valgrind is sigKILL'd on startup 2080117295 == 117290 2081118703 m_signals.c:1427 Assertion 'tst->status == VgTs_WaitSys' 2082118466 add %reg, %reg generates incorrect validity for bit 0 2083123210 New: strlen from ld-linux on amd64 2084123244 DWARF2 CFI reader: unhandled CFI instruction 0:18 2085123248 syscalls in glibc-2.4: openat, fstatat, symlinkat 2086123258 socketcall.recvmsg(msg.msg_iov[i] points to uninit 2087123535 mremap(new_addr) requires MREMAP_FIXED in 4th arg 2088123836 small typo in the doc 2089124029 ppc compile failed: `vor' gcc 3.3.5 2090124222 Segfault: @@don't know what type ':' is 2091124475 ppc32: crash (syscall?) timer_settime() 2092124499 amd64->IR: 0xF 0xE 0x48 0x85 (femms) 2093124528 FATAL: aspacem assertion failed: segment_is_sane 2094124697 vex x86->IR: 0xF 0x70 0xC9 0x0 (pshufw) 2095124892 vex x86->IR: 0xF3 0xAE (REPx SCASB) 2096126216 == 124892 2097124808 ppc32: sys_sched_getaffinity() not handled 2098n-i-bz Very long stabs strings crash m_debuginfo 2099n-i-bz amd64->IR: 0x66 0xF 0xF5 (pmaddwd) 2100125492 ppc32: support a bunch more syscalls 2101121617 ppc32/64: coredumping gives assertion failure 2102121814 Coregrind return error as exitcode patch 2103126517 == 121814 2104125607 amd64->IR: 0x66 0xF 0xA3 0x2 (btw etc) 2105125651 amd64->IR: 0xF8 0x49 0xFF 0xE3 (clc?) 2106126253 x86 movx is wrong 2107126451 3.2 SVN doesn't work on ppc32 CPU's without FPU 2108126217 increase # threads 2109126243 vex x86->IR: popw mem 2110126583 amd64->IR: 0x48 0xF 0xA4 0xC2 (shld $1,%rax,%rdx) 2111126668 amd64->IR: 0x1C 0xFF (sbb $0xff,%al) 2112126696 support for CDROMREADRAW ioctl and CDROMREADTOCENTRY fix 2113126722 assertion: segment_is_sane at m_aspacemgr/aspacemgr.c:1624 2114126938 bad checking for syscalls linkat, renameat, symlinkat 2115 2116(3.2.0RC1: 27 May 2006, vex r1626, valgrind r5947). 2117(3.2.0: 7 June 2006, vex r1628, valgrind r5957). 2118 2119 2120Release 3.1.1 (15 March 2006) 2121~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 21223.1.1 fixes a bunch of bugs reported in 3.1.0. There is no new 2123functionality. The fixed bugs are: 2124 2125(note: "n-i-bz" means "not in bugzilla" -- this bug does not have 2126 a bugzilla entry). 2127 2128n-i-bz ppc32: fsub 3,3,3 in dispatcher doesn't clear NaNs 2129n-i-bz ppc32: __NR_{set,get}priority 2130117332 x86: missing line info with icc 8.1 2131117366 amd64: 0xDD 0x7C fnstsw 2132118274 == 117366 2133117367 amd64: 0xD9 0xF4 fxtract 2134117369 amd64: __NR_getpriority (140) 2135117419 ppc32: lfsu f5, -4(r11) 2136117419 ppc32: fsqrt 2137117936 more stabs problems (segfaults while reading debug info) 2138119914 == 117936 2139120345 == 117936 2140118239 amd64: 0xF 0xAE 0x3F (clflush) 2141118939 vm86old system call 2142n-i-bz memcheck/tests/mempool reads freed memory 2143n-i-bz AshleyP's custom-allocator assertion 2144n-i-bz Dirk strict-aliasing stuff 2145n-i-bz More space for debugger cmd line (Dan Thaler) 2146n-i-bz Clarified leak checker output message 2147n-i-bz AshleyP's --gen-suppressions output fix 2148n-i-bz cg_annotate's --sort option broken 2149n-i-bz OSet 64-bit fastcmp bug 2150n-i-bz VG_(getgroups) fix (Shinichi Noda) 2151n-i-bz ppc32: allocate from callee-saved FP/VMX regs 2152n-i-bz misaligned path word-size bug in mc_main.c 2153119297 Incorrect error message for sse code 2154120410 x86: prefetchw (0xF 0xD 0x48 0x4) 2155120728 TIOCSERGETLSR, TIOCGICOUNT, HDIO_GET_DMA ioctls 2156120658 Build fixes for gcc 2.96 2157120734 x86: Support for changing EIP in signal handler 2158n-i-bz memcheck/tests/zeropage de-looping fix 2159n-i-bz x86: fxtract doesn't work reliably 2160121662 x86: lock xadd (0xF0 0xF 0xC0 0x2) 2161121893 calloc does not always return zeroed memory 2162121901 no support for syscall tkill 2163n-i-bz Suppression update for Debian unstable 2164122067 amd64: fcmovnu (0xDB 0xD9) 2165n-i-bz ppc32: broken signal handling in cpu feature detection 2166n-i-bz ppc32: rounding mode problems (improved, partial fix only) 2167119482 ppc32: mtfsb1 2168n-i-bz ppc32: mtocrf/mfocrf 2169 2170(3.1.1: 15 March 2006, vex r1597, valgrind r5771). 2171 2172 2173Release 3.1.0 (25 November 2005) 2174~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 21753.1.0 is a feature release with a number of significant improvements: 2176AMD64 support is much improved, PPC32 support is good enough to be 2177usable, and the handling of memory management and address space is 2178much more robust. In detail: 2179 2180- AMD64 support is much improved. The 64-bit vs. 32-bit issues in 2181 3.0.X have been resolved, and it should "just work" now in all 2182 cases. On AMD64 machines both 64-bit and 32-bit versions of 2183 Valgrind are built. The right version will be invoked 2184 automatically, even when using --trace-children and mixing execution 2185 between 64-bit and 32-bit executables. Also, many more instructions 2186 are supported. 2187 2188- PPC32 support is now good enough to be usable. It should work with 2189 all tools, but please let us know if you have problems. Three 2190 classes of CPUs are supported: integer only (no FP, no Altivec), 2191 which covers embedded PPC uses, integer and FP but no Altivec 2192 (G3-ish), and CPUs capable of Altivec too (G4, G5). 2193 2194- Valgrind's address space management has been overhauled. As a 2195 result, Valgrind should be much more robust with programs that use 2196 large amounts of memory. There should be many fewer "memory 2197 exhausted" messages, and debug symbols should be read correctly on 2198 large (eg. 300MB+) executables. On 32-bit machines the full address 2199 space available to user programs (usually 3GB or 4GB) can be fully 2200 utilised. On 64-bit machines up to 32GB of space is usable; when 2201 using Memcheck that means your program can use up to about 14GB. 2202 2203 A side effect of this change is that Valgrind is no longer protected 2204 against wild writes by the client. This feature was nice but relied 2205 on the x86 segment registers and so wasn't portable. 2206 2207- Most users should not notice, but as part of the address space 2208 manager change, the way Valgrind is built has been changed. Each 2209 tool is now built as a statically linked stand-alone executable, 2210 rather than as a shared object that is dynamically linked with the 2211 core. The "valgrind" program invokes the appropriate tool depending 2212 on the --tool option. This slightly increases the amount of disk 2213 space used by Valgrind, but it greatly simplified many things and 2214 removed Valgrind's dependence on glibc. 2215 2216Please note that Addrcheck and Helgrind are still not working. Work 2217is underway to reinstate them (or equivalents). We apologise for the 2218inconvenience. 2219 2220Other user-visible changes: 2221 2222- The --weird-hacks option has been renamed --sim-hints. 2223 2224- The --time-stamp option no longer gives an absolute date and time. 2225 It now prints the time elapsed since the program began. 2226 2227- It should build with gcc-2.96. 2228 2229- Valgrind can now run itself (see README_DEVELOPERS for how). 2230 This is not much use to you, but it means the developers can now 2231 profile Valgrind using Cachegrind. As a result a couple of 2232 performance bad cases have been fixed. 2233 2234- The XML output format has changed slightly. See 2235 docs/internals/xml-output.txt. 2236 2237- Core dumping has been reinstated (it was disabled in 3.0.0 and 3.0.1). 2238 If your program crashes while running under Valgrind, a core file with 2239 the name "vgcore.<pid>" will be created (if your settings allow core 2240 file creation). Note that the floating point information is not all 2241 there. If Valgrind itself crashes, the OS will create a normal core 2242 file. 2243 2244The following are some user-visible changes that occurred in earlier 2245versions that may not have been announced, or were announced but not 2246widely noticed. So we're mentioning them now. 2247 2248- The --tool flag is optional once again; if you omit it, Memcheck 2249 is run by default. 2250 2251- The --num-callers flag now has a default value of 12. It was 2252 previously 4. 2253 2254- The --xml=yes flag causes Valgrind's output to be produced in XML 2255 format. This is designed to make it easy for other programs to 2256 consume Valgrind's output. The format is described in the file 2257 docs/internals/xml-format.txt. 2258 2259- The --gen-suppressions flag supports an "all" value that causes every 2260 suppression to be printed without asking. 2261 2262- The --log-file option no longer puts "pid" in the filename, eg. the 2263 old name "foo.pid12345" is now "foo.12345". 2264 2265- There are several graphical front-ends for Valgrind, such as Valkyrie, 2266 Alleyoop and Valgui. See http://www.valgrind.org/downloads/guis.html 2267 for a list. 2268 2269BUGS FIXED: 2270 2271109861 amd64 hangs at startup 2272110301 ditto 2273111554 valgrind crashes with Cannot allocate memory 2274111809 Memcheck tool doesn't start java 2275111901 cross-platform run of cachegrind fails on opteron 2276113468 (vgPlain_mprotect_range): Assertion 'r != -1' failed. 2277 92071 Reading debugging info uses too much memory 2278109744 memcheck loses track of mmap from direct ld-linux.so.2 2279110183 tail of page with _end 2280 82301 FV memory layout too rigid 2281 98278 Infinite recursion possible when allocating memory 2282108994 Valgrind runs out of memory due to 133x overhead 2283115643 valgrind cannot allocate memory 2284105974 vg_hashtable.c static hash table 2285109323 ppc32: dispatch.S uses Altivec insn, which doesn't work on POWER. 2286109345 ptrace_setregs not yet implemented for ppc 2287110831 Would like to be able to run against both 32 and 64 bit 2288 binaries on AMD64 2289110829 == 110831 2290111781 compile of valgrind-3.0.0 fails on my linux (gcc 2.X prob) 2291112670 Cachegrind: cg_main.c:486 (handleOneStatement ... 2292112941 vex x86: 0xD9 0xF4 (fxtract) 2293110201 == 112941 2294113015 vex amd64->IR: 0xE3 0x14 0x48 0x83 (jrcxz) 2295113126 Crash with binaries built with -gstabs+/-ggdb 2296104065 == 113126 2297115741 == 113126 2298113403 Partial SSE3 support on x86 2299113541 vex: Grp5(x86) (alt encoding inc/dec) case 1 2300113642 valgrind crashes when trying to read debug information 2301113810 vex x86->IR: 66 0F F6 (66 + PSADBW == SSE PSADBW) 2302113796 read() and write() do not work if buffer is in shared memory 2303113851 vex x86->IR: (pmaddwd): 0x66 0xF 0xF5 0xC7 2304114366 vex amd64 cannnot handle __asm__( "fninit" ) 2305114412 vex amd64->IR: 0xF 0xAD 0xC2 0xD3 (128-bit shift, shrdq?) 2306114455 vex amd64->IR: 0xF 0xAC 0xD0 0x1 (also shrdq) 2307115590: amd64->IR: 0x67 0xE3 0x9 0xEB (address size override) 2308115953 valgrind svn r5042 does not build with parallel make (-j3) 2309116057 maximum instruction size - VG_MAX_INSTR_SZB too small? 2310116483 shmat failes with invalid argument 2311102202 valgrind crashes when realloc'ing until out of memory 2312109487 == 102202 2313110536 == 102202 2314112687 == 102202 2315111724 vex amd64->IR: 0x41 0xF 0xAB (more BT{,S,R,C} fun n games) 2316111748 vex amd64->IR: 0xDD 0xE2 (fucom) 2317111785 make fails if CC contains spaces 2318111829 vex x86->IR: sbb AL, Ib 2319111851 vex x86->IR: 0x9F 0x89 (lahf/sahf) 2320112031 iopl on AMD64 and README_MISSING_SYSCALL_OR_IOCTL update 2321112152 code generation for Xin_MFence on x86 with SSE0 subarch 2322112167 == 112152 2323112789 == 112152 2324112199 naked ar tool is used in vex makefile 2325112501 vex x86->IR: movq (0xF 0x7F 0xC1 0xF) (mmx MOVQ) 2326113583 == 112501 2327112538 memalign crash 2328113190 Broken links in docs/html/ 2329113230 Valgrind sys_pipe on x86-64 wrongly thinks file descriptors 2330 should be 64bit 2331113996 vex amd64->IR: fucomp (0xDD 0xE9) 2332114196 vex x86->IR: out %eax,(%dx) (0xEF 0xC9 0xC3 0x90) 2333114289 Memcheck fails to intercept malloc when used in an uclibc environment 2334114756 mbind syscall support 2335114757 Valgrind dies with assertion: Assertion 'noLargerThan > 0' failed 2336114563 stack tracking module not informed when valgrind switches threads 2337114564 clone() and stacks 2338114565 == 114564 2339115496 glibc crashes trying to use sysinfo page 2340116200 enable fsetxattr, fgetxattr, and fremovexattr for amd64 2341 2342(3.1.0RC1: 20 November 2005, vex r1466, valgrind r5224). 2343(3.1.0: 26 November 2005, vex r1471, valgrind r5235). 2344 2345 2346Release 3.0.1 (29 August 2005) 2347~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 23483.0.1 fixes a bunch of bugs reported in 3.0.0. There is no new 2349functionality. Some of the fixed bugs are critical, so if you 2350use/distribute 3.0.0, an upgrade to 3.0.1 is recommended. The fixed 2351bugs are: 2352 2353(note: "n-i-bz" means "not in bugzilla" -- this bug does not have 2354 a bugzilla entry). 2355 2356109313 (== 110505) x86 cmpxchg8b 2357n-i-bz x86: track but ignore changes to %eflags.AC (alignment check) 2358110102 dis_op2_E_G(amd64) 2359110202 x86 sys_waitpid(#286) 2360110203 clock_getres(,0) 2361110208 execve fail wrong retval 2362110274 SSE1 now mandatory for x86 2363110388 amd64 0xDD 0xD1 2364110464 amd64 0xDC 0x1D FCOMP 2365110478 amd64 0xF 0xD PREFETCH 2366n-i-bz XML <unique> printing wrong 2367n-i-bz Dirk r4359 (amd64 syscalls from trunk) 2368110591 amd64 and x86: rdtsc not implemented properly 2369n-i-bz Nick r4384 (stub implementations of Addrcheck and Helgrind) 2370110652 AMD64 valgrind crashes on cwtd instruction 2371110653 AMD64 valgrind crashes on sarb $0x4,foo(%rip) instruction 2372110656 PATH=/usr/bin::/bin valgrind foobar stats ./fooba 2373110657 Small test fixes 2374110671 vex x86->IR: unhandled instruction bytes: 0xF3 0xC3 (rep ret) 2375n-i-bz Nick (Cachegrind should not assert when it encounters a client 2376 request.) 2377110685 amd64->IR: unhandled instruction bytes: 0xE1 0x56 (loope Jb) 2378110830 configuring with --host fails to build 32 bit on 64 bit target 2379110875 Assertion when execve fails 2380n-i-bz Updates to Memcheck manual 2381n-i-bz Fixed broken malloc_usable_size() 2382110898 opteron instructions missing: btq btsq btrq bsfq 2383110954 x86->IR: unhandled instruction bytes: 0xE2 0xF6 (loop Jb) 2384n-i-bz Make suppressions work for "???" lines in stacktraces. 2385111006 bogus warnings from linuxthreads 2386111092 x86: dis_Grp2(Reg): unhandled case(x86) 2387111231 sctp_getladdrs() and sctp_getpaddrs() returns uninitialized 2388 memory 2389111102 (comment #4) Fixed 64-bit unclean "silly arg" message 2390n-i-bz vex x86->IR: unhandled instruction bytes: 0x14 0x0 2391n-i-bz minor umount/fcntl wrapper fixes 2392111090 Internal Error running Massif 2393101204 noisy warning 2394111513 Illegal opcode for SSE instruction (x86 movups) 2395111555 VEX/Makefile: CC is set to gcc 2396n-i-bz Fix XML bugs in FAQ 2397 2398(3.0.1: 29 August 05, 2399 vex/branches/VEX_3_0_BRANCH r1367, 2400 valgrind/branches/VALGRIND_3_0_BRANCH r4574). 2401 2402 2403 2404Release 3.0.0 (3 August 2005) 2405~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 24063.0.0 is a major overhaul of Valgrind. The most significant user 2407visible change is that Valgrind now supports architectures other than 2408x86. The new architectures it supports are AMD64 and PPC32, and the 2409infrastructure is present for other architectures to be added later. 2410 2411AMD64 support works well, but has some shortcomings: 2412 2413- It generally won't be as solid as the x86 version. For example, 2414 support for more obscure instructions and system calls may be missing. 2415 We will fix these as they arise. 2416 2417- Address space may be limited; see the point about 2418 position-independent executables below. 2419 2420- If Valgrind is built on an AMD64 machine, it will only run 64-bit 2421 executables. If you want to run 32-bit x86 executables under Valgrind 2422 on an AMD64, you will need to build Valgrind on an x86 machine and 2423 copy it to the AMD64 machine. And it probably won't work if you do 2424 something tricky like exec'ing a 32-bit program from a 64-bit program 2425 while using --trace-children=yes. We hope to improve this situation 2426 in the future. 2427 2428The PPC32 support is very basic. It may not work reliably even for 2429small programs, but it's a start. Many thanks to Paul Mackerras for 2430his great work that enabled this support. We are working to make 2431PPC32 usable as soon as possible. 2432 2433Other user-visible changes: 2434 2435- Valgrind is no longer built by default as a position-independent 2436 executable (PIE), as this caused too many problems. 2437 2438 Without PIE enabled, AMD64 programs will only be able to access 2GB of 2439 address space. We will fix this eventually, but not for the moment. 2440 2441 Use --enable-pie at configure-time to turn this on. 2442 2443- Support for programs that use stack-switching has been improved. Use 2444 the --max-stackframe flag for simple cases, and the 2445 VALGRIND_STACK_REGISTER, VALGRIND_STACK_DEREGISTER and 2446 VALGRIND_STACK_CHANGE client requests for trickier cases. 2447 2448- Support for programs that use self-modifying code has been improved, 2449 in particular programs that put temporary code fragments on the stack. 2450 This helps for C programs compiled with GCC that use nested functions, 2451 and also Ada programs. This is controlled with the --smc-check 2452 flag, although the default setting should work in most cases. 2453 2454- Output can now be printed in XML format. This should make it easier 2455 for tools such as GUI front-ends and automated error-processing 2456 schemes to use Valgrind output as input. The --xml flag controls this. 2457 As part of this change, ELF directory information is read from executables, 2458 so absolute source file paths are available if needed. 2459 2460- Programs that allocate many heap blocks may run faster, due to 2461 improvements in certain data structures. 2462 2463- Addrcheck is currently not working. We hope to get it working again 2464 soon. Helgrind is still not working, as was the case for the 2.4.0 2465 release. 2466 2467- The JITter has been completely rewritten, and is now in a separate 2468 library, called Vex. This enabled a lot of the user-visible changes, 2469 such as new architecture support. The new JIT unfortunately translates 2470 more slowly than the old one, so programs may take longer to start. 2471 We believe the code quality is produces is about the same, so once 2472 started, programs should run at about the same speed. Feedback about 2473 this would be useful. 2474 2475 On the plus side, Vex and hence Memcheck tracks value flow properly 2476 through floating point and vector registers, something the 2.X line 2477 could not do. That means that Memcheck is much more likely to be 2478 usably accurate on vectorised code. 2479 2480- There is a subtle change to the way exiting of threaded programs 2481 is handled. In 3.0, Valgrind's final diagnostic output (leak check, 2482 etc) is not printed until the last thread exits. If the last thread 2483 to exit was not the original thread which started the program, any 2484 other process wait()-ing on this one to exit may conclude it has 2485 finished before the diagnostic output is printed. This may not be 2486 what you expect. 2.X had a different scheme which avoided this 2487 problem, but caused deadlocks under obscure circumstances, so we 2488 are trying something different for 3.0. 2489 2490- Small changes in control log file naming which make it easier to 2491 use valgrind for debugging MPI-based programs. The relevant 2492 new flags are --log-file-exactly= and --log-file-qualifier=. 2493 2494- As part of adding AMD64 support, DWARF2 CFI-based stack unwinding 2495 support was added. In principle this means Valgrind can produce 2496 meaningful backtraces on x86 code compiled with -fomit-frame-pointer 2497 providing you also compile your code with -fasynchronous-unwind-tables. 2498 2499- The documentation build system has been completely redone. 2500 The documentation masters are now in XML format, and from that 2501 HTML, PostScript and PDF documentation is generated. As a result 2502 the manual is now available in book form. Note that the 2503 documentation in the source tarballs is pre-built, so you don't need 2504 any XML processing tools to build Valgrind from a tarball. 2505 2506Changes that are not user-visible: 2507 2508- The code has been massively overhauled in order to modularise it. 2509 As a result we hope it is easier to navigate and understand. 2510 2511- Lots of code has been rewritten. 2512 2513BUGS FIXED: 2514 2515110046 sz == 4 assertion failed 2516109810 vex amd64->IR: unhandled instruction bytes: 0xA3 0x4C 0x70 0xD7 2517109802 Add a plausible_stack_size command-line parameter ? 2518109783 unhandled ioctl TIOCMGET (running hw detection tool discover) 2519109780 unhandled ioctl BLKSSZGET (running fdisk -l /dev/hda) 2520109718 vex x86->IR: unhandled instruction: ffreep 2521109429 AMD64 unhandled syscall: 127 (sigpending) 2522109401 false positive uninit in strchr from ld-linux.so.2 2523109385 "stabs" parse failure 2524109378 amd64: unhandled instruction REP NOP 2525109376 amd64: unhandled instruction LOOP Jb 2526109363 AMD64 unhandled instruction bytes 2527109362 AMD64 unhandled syscall: 24 (sched_yield) 2528109358 fork() won't work with valgrind-3.0 SVN 2529109332 amd64 unhandled instruction: ADC Ev, Gv 2530109314 Bogus memcheck report on amd64 2531108883 Crash; vg_memory.c:905 (vgPlain_init_shadow_range): 2532 Assertion `vgPlain_defined_init_shadow_page()' failed. 2533108349 mincore syscall parameter checked incorrectly 2534108059 build infrastructure: small update 2535107524 epoll_ctl event parameter checked on EPOLL_CTL_DEL 2536107123 Vex dies with unhandled instructions: 0xD9 0x31 0xF 0xAE 2537106841 auxmap & openGL problems 2538106713 SDL_Init causes valgrind to exit 2539106352 setcontext and makecontext not handled correctly 2540106293 addresses beyond initial client stack allocation 2541 not checked in VALGRIND_DO_LEAK_CHECK 2542106283 PIE client programs are loaded at address 0 2543105831 Assertion `vgPlain_defined_init_shadow_page()' failed. 2544105039 long run-times probably due to memory manager 2545104797 valgrind needs to be aware of BLKGETSIZE64 2546103594 unhandled instruction: FICOM 2547103320 Valgrind 2.4.0 fails to compile with gcc 3.4.3 and -O0 2548103168 potentially memory leak in coregrind/ume.c 2549102039 bad permissions for mapped region at address 0xB7C73680 2550101881 weird assertion problem 2551101543 Support fadvise64 syscalls 255275247 x86_64/amd64 support (the biggest "bug" we have ever fixed) 2553 2554(3.0RC1: 27 July 05, vex r1303, valgrind r4283). 2555(3.0.0: 3 August 05, vex r1313, valgrind r4316). 2556 2557 2558 2559Stable release 2.4.1 (1 August 2005) 2560~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2561(The notes for this release have been lost. Sorry! It would have 2562contained various bug fixes but no new features.) 2563 2564 2565 2566Stable release 2.4.0 (March 2005) -- CHANGES RELATIVE TO 2.2.0 2567~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 25682.4.0 brings many significant changes and bug fixes. The most 2569significant user-visible change is that we no longer supply our own 2570pthread implementation. Instead, Valgrind is finally capable of 2571running the native thread library, either LinuxThreads or NPTL. 2572 2573This means our libpthread has gone, along with the bugs associated 2574with it. Valgrind now supports the kernel's threading syscalls, and 2575lets you use your standard system libpthread. As a result: 2576 2577* There are many fewer system dependencies and strange library-related 2578 bugs. There is a small performance improvement, and a large 2579 stability improvement. 2580 2581* On the downside, Valgrind can no longer report misuses of the POSIX 2582 PThreads API. It also means that Helgrind currently does not work. 2583 We hope to fix these problems in a future release. 2584 2585Note that running the native thread libraries does not mean Valgrind 2586is able to provide genuine concurrent execution on SMPs. We still 2587impose the restriction that only one thread is running at any given 2588time. 2589 2590There are many other significant changes too: 2591 2592* Memcheck is (once again) the default tool. 2593 2594* The default stack backtrace is now 12 call frames, rather than 4. 2595 2596* Suppressions can have up to 25 call frame matches, rather than 4. 2597 2598* Memcheck and Addrcheck use less memory. Under some circumstances, 2599 they no longer allocate shadow memory if there are large regions of 2600 memory with the same A/V states - such as an mmaped file. 2601 2602* The memory-leak detector in Memcheck and Addrcheck has been 2603 improved. It now reports more types of memory leak, including 2604 leaked cycles. When reporting leaked memory, it can distinguish 2605 between directly leaked memory (memory with no references), and 2606 indirectly leaked memory (memory only referred to by other leaked 2607 memory). 2608 2609* Memcheck's confusion over the effect of mprotect() has been fixed: 2610 previously mprotect could erroneously mark undefined data as 2611 defined. 2612 2613* Signal handling is much improved and should be very close to what 2614 you get when running natively. 2615 2616 One result of this is that Valgrind observes changes to sigcontexts 2617 passed to signal handlers. Such modifications will take effect when 2618 the signal returns. You will need to run with --single-step=yes to 2619 make this useful. 2620 2621* Valgrind is built in Position Independent Executable (PIE) format if 2622 your toolchain supports it. This allows it to take advantage of all 2623 the available address space on systems with 4Gbyte user address 2624 spaces. 2625 2626* Valgrind can now run itself (requires PIE support). 2627 2628* Syscall arguments are now checked for validity. Previously all 2629 memory used by syscalls was checked, but now the actual values 2630 passed are also checked. 2631 2632* Syscall wrappers are more robust against bad addresses being passed 2633 to syscalls: they will fail with EFAULT rather than killing Valgrind 2634 with SIGSEGV. 2635 2636* Because clone() is directly supported, some non-pthread uses of it 2637 will work. Partial sharing (where some resources are shared, and 2638 some are not) is not supported. 2639 2640* open() and readlink() on /proc/self/exe are supported. 2641 2642BUGS FIXED: 2643 264488520 pipe+fork+dup2 kills the main program 264588604 Valgrind Aborts when using $VALGRIND_OPTS and user progra... 264688614 valgrind: vg_libpthread.c:2323 (read): Assertion `read_pt... 264788703 Stabs parser fails to handle ";" 264888886 ioctl wrappers for TIOCMBIS and TIOCMBIC 264989032 valgrind pthread_cond_timedwait fails 265089106 the 'impossible' happened 265189139 Missing sched_setaffinity & sched_getaffinity 265289198 valgrind lacks support for SIOCSPGRP and SIOCGPGRP 265389263 Missing ioctl translations for scsi-generic and CD playing 265489440 tests/deadlock.c line endings 265589481 `impossible' happened: EXEC FAILED 265689663 valgrind 2.2.0 crash on Redhat 7.2 265789792 Report pthread_mutex_lock() deadlocks instead of returnin... 265890111 statvfs64 gives invalid error/warning 265990128 crash+memory fault with stabs generated by gnat for a run... 266090778 VALGRIND_CHECK_DEFINED() not as documented in memcheck.h 266190834 cachegrind crashes at end of program without reporting re... 266291028 valgrind: vg_memory.c:229 (vgPlain_unmap_range): Assertio... 266391162 valgrind crash while debugging drivel 1.2.1 266491199 Unimplemented function 266591325 Signal routing does not propagate the siginfo structure 266691599 Assertion `cv == ((void *)0)' 266791604 rw_lookup clears orig and sends the NULL value to rw_new 266891821 Small problems building valgrind with $top_builddir ne $t... 266991844 signal 11 (SIGSEGV) at get_tcb (libpthread.c:86) in corec... 267092264 UNIMPLEMENTED FUNCTION: pthread_condattr_setpshared 267192331 per-target flags necessitate AM_PROG_CC_C_O 267292420 valgrind doesn't compile with linux 2.6.8.1/9 267392513 Valgrind 2.2.0 generates some warning messages 267492528 vg_symtab2.c:170 (addLoc): Assertion `loc->size > 0' failed. 267593096 unhandled ioctl 0x4B3A and 0x5601 267693117 Tool and core interface versions do not match 267793128 Can't run valgrind --tool=memcheck because of unimplement... 267893174 Valgrind can crash if passed bad args to certain syscalls 267993309 Stack frame in new thread is badly aligned 268093328 Wrong types used with sys_sigprocmask() 268193763 /usr/include/asm/msr.h is missing 268293776 valgrind: vg_memory.c:508 (vgPlain_find_map_space): Asser... 268393810 fcntl() argument checking a bit too strict 268494378 Assertion `tst->sigqueue_head != tst->sigqueue_tail' failed. 268594429 valgrind 2.2.0 segfault with mmap64 in glibc 2.3.3 268694645 Impossible happened: PINSRW mem 268794953 valgrind: the `impossible' happened: SIGSEGV 268895667 Valgrind does not work with any KDE app 268996243 Assertion 'res==0' failed 269096252 stage2 loader of valgrind fails to allocate memory 269196520 All programs crashing at _dl_start (in /lib/ld-2.3.3.so) ... 269296660 ioctl CDROMREADTOCENTRY causes bogus warnings 269396747 After looping in a segfault handler, the impossible happens 269496923 Zero sized arrays crash valgrind trace back with SIGFPE 269596948 valgrind stops with assertion failure regarding mmap2 269696966 valgrind fails when application opens more than 16 sockets 269797398 valgrind: vg_libpthread.c:2667 Assertion failed 269897407 valgrind: vg_mylibc.c:1226 (vgPlain_safe_fd): Assertion `... 269997427 "Warning: invalid file descriptor -1 in syscall close()" ... 270097785 missing backtrace 270197792 build in obj dir fails - autoconf / makefile cleanup 270297880 pthread_mutex_lock fails from shared library (special ker... 270397975 program aborts without ang VG messages 270498129 Failed when open and close file 230000 times using stdio 270598175 Crashes when using valgrind-2.2.0 with a program using al... 270698288 Massif broken 270798303 UNIMPLEMENTED FUNCTION pthread_condattr_setpshared 270898630 failed--compilation missing warnings.pm, fails to make he... 270998756 Cannot valgrind signal-heavy kdrive X server 271098966 valgrinding the JVM fails with a sanity check assertion 271199035 Valgrind crashes while profiling 271299142 loops with message "Signal 11 being dropped from thread 0... 271399195 threaded apps crash on thread start (using QThread::start... 271499348 Assertion `vgPlain_lseek(core_fd, 0, 1) == phdrs[i].p_off... 271599568 False negative due to mishandling of mprotect 271699738 valgrind memcheck crashes on program that uses sigitimer 271799923 0-sized allocations are reported as leaks 271899949 program seg faults after exit() 2719100036 "newSuperblock's request for 1048576 bytes failed" 2720100116 valgrind: (pthread_cond_init): Assertion `sizeof(* cond) ... 2721100486 memcheck reports "valgrind: the `impossible' happened: V... 2722100833 second call to "mremap" fails with EINVAL 2723101156 (vgPlain_find_map_space): Assertion `(addr & ((1 << 12)-1... 2724101173 Assertion `recDepth >= 0 && recDepth < 500' failed 2725101291 creating threads in a forked process fails 2726101313 valgrind causes different behavior when resizing a window... 2727101423 segfault for c++ array of floats 2728101562 valgrind massif dies on SIGINT even with signal handler r... 2729 2730 2731Stable release 2.2.0 (31 August 2004) -- CHANGES RELATIVE TO 2.0.0 2732~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 27332.2.0 brings nine months worth of improvements and bug fixes. We 2734believe it to be a worthy successor to 2.0.0. There are literally 2735hundreds of bug fixes and minor improvements. There are also some 2736fairly major user-visible changes: 2737 2738* A complete overhaul of handling of system calls and signals, and 2739 their interaction with threads. In general, the accuracy of the 2740 system call, thread and signal simulations is much improved: 2741 2742 - Blocking system calls behave exactly as they do when running 2743 natively (not on valgrind). That is, if a syscall blocks only the 2744 calling thread when running natively, than it behaves the same on 2745 valgrind. No more mysterious hangs because V doesn't know that some 2746 syscall or other, should block only the calling thread. 2747 2748 - Interrupted syscalls should now give more faithful results. 2749 2750 - Signal contexts in signal handlers are supported. 2751 2752* Improvements to NPTL support to the extent that V now works 2753 properly on NPTL-only setups. 2754 2755* Greater isolation between Valgrind and the program being run, so 2756 the program is less likely to inadvertently kill Valgrind by 2757 doing wild writes. 2758 2759* Massif: a new space profiling tool. Try it! It's cool, and it'll 2760 tell you in detail where and when your C/C++ code is allocating heap. 2761 Draws pretty .ps pictures of memory use against time. A potentially 2762 powerful tool for making sense of your program's space use. 2763 2764* File descriptor leakage checks. When enabled, Valgrind will print out 2765 a list of open file descriptors on exit. 2766 2767* Improved SSE2/SSE3 support. 2768 2769* Time-stamped output; use --time-stamp=yes 2770 2771 2772 2773Stable release 2.2.0 (31 August 2004) -- CHANGES RELATIVE TO 2.1.2 2774~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 27752.2.0 is not much different from 2.1.2, released seven weeks ago. 2776A number of bugs have been fixed, most notably #85658, which gave 2777problems for quite a few people. There have been many internal 2778cleanups, but those are not user visible. 2779 2780The following bugs have been fixed since 2.1.2: 2781 278285658 Assert in coregrind/vg_libpthread.c:2326 (open64) != 2783 (void*)0 failed 2784 This bug was reported multiple times, and so the following 2785 duplicates of it are also fixed: 87620, 85796, 85935, 86065, 2786 86919, 86988, 87917, 88156 2787 278880716 Semaphore mapping bug caused by unmap (sem_destroy) 2789 (Was fixed prior to 2.1.2) 2790 279186987 semctl and shmctl syscalls family is not handled properly 2792 279386696 valgrind 2.1.2 + RH AS2.1 + librt 2794 279586730 valgrind locks up at end of run with assertion failure 2796 in __pthread_unwind 2797 279886641 memcheck doesn't work with Mesa OpenGL/ATI on Suse 9.1 2799 (also fixes 74298, a duplicate of this) 2800 280185947 MMX/SSE unhandled instruction 'sfence' 2802 280384978 Wrong error "Conditional jump or move depends on 2804 uninitialised value" resulting from "sbbl %reg, %reg" 2805 280686254 ssort() fails when signed int return type from comparison is 2807 too small to handle result of unsigned int subtraction 2808 280987089 memalign( 4, xxx) makes valgrind assert 2810 281186407 Add support for low-level parallel port driver ioctls. 2812 281370587 Add timestamps to Valgrind output? (wishlist) 2814 281584937 vg_libpthread.c:2505 (se_remap): Assertion `res == 0' 2816 (fixed prior to 2.1.2) 2817 281886317 cannot load libSDL-1.2.so.0 using valgrind 2819 282086989 memcpy from mac_replace_strmem.c complains about 2821 uninitialized pointers passed when length to copy is zero 2822 282385811 gnu pascal symbol causes segmentation fault; ok in 2.0.0 2824 282579138 writing to sbrk()'d memory causes segfault 2826 282777369 sched deadlock while signal received during pthread_join 2828 and the joined thread exited 2829 283088115 In signal handler for SIGFPE, siginfo->si_addr is wrong 2831 under Valgrind 2832 283378765 Massif crashes on app exit if FP exceptions are enabled 2834 2835Additionally there are the following changes, which are not 2836connected to any bug report numbers, AFAICS: 2837 2838* Fix scary bug causing mis-identification of SSE stores vs 2839 loads and so causing memcheck to sometimes give nonsense results 2840 on SSE code. 2841 2842* Add support for the POSIX message queue system calls. 2843 2844* Fix to allow 32-bit Valgrind to run on AMD64 boxes. Note: this does 2845 NOT allow Valgrind to work with 64-bit executables - only with 32-bit 2846 executables on an AMD64 box. 2847 2848* At configure time, only check whether linux/mii.h can be processed 2849 so that we don't generate ugly warnings by trying to compile it. 2850 2851* Add support for POSIX clocks and timers. 2852 2853 2854 2855Developer (cvs head) release 2.1.2 (18 July 2004) 2856~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 28572.1.2 contains four months worth of bug fixes and refinements. 2858Although officially a developer release, we believe it to be stable 2859enough for widespread day-to-day use. 2.1.2 is pretty good, so try it 2860first, although there is a chance it won't work. If so then try 2.0.0 2861and tell us what went wrong." 2.1.2 fixes a lot of problems present 2862in 2.0.0 and is generally a much better product. 2863 2864Relative to 2.1.1, a large number of minor problems with 2.1.1 have 2865been fixed, and so if you use 2.1.1 you should try 2.1.2. Users of 2866the last stable release, 2.0.0, might also want to try this release. 2867 2868The following bugs, and probably many more, have been fixed. These 2869are listed at http://bugs.kde.org. Reporting a bug for valgrind in 2870the http://bugs.kde.org is much more likely to get you a fix than 2871mailing developers directly, so please continue to keep sending bugs 2872there. 2873 287476869 Crashes when running any tool under Fedora Core 2 test1 2875 This fixes the problem with returning from a signal handler 2876 when VDSOs are turned off in FC2. 2877 287869508 java 1.4.2 client fails with erroneous "stack size too small". 2879 This fix makes more of the pthread stack attribute related 2880 functions work properly. Java still doesn't work though. 2881 288271906 malloc alignment should be 8, not 4 2883 All memory returned by malloc/new etc is now at least 2884 8-byte aligned. 2885 288681970 vg_alloc_ThreadState: no free slots available 2887 (closed because the workaround is simple: increase 2888 VG_N_THREADS, rebuild and try again.) 2889 289078514 Conditional jump or move depends on uninitialized value(s) 2891 (a slight mishanding of FP code in memcheck) 2892 289377952 pThread Support (crash) (due to initialisation-ordering probs) 2894 (also 85118) 2895 289680942 Addrcheck wasn't doing overlap checking as it should. 289778048 return NULL on malloc/new etc failure, instead of asserting 289873655 operator new() override in user .so files often doesn't get picked up 289983060 Valgrind does not handle native kernel AIO 290069872 Create proper coredumps after fatal signals 290182026 failure with new glibc versions: __libc_* functions are not exported 290270344 UNIMPLEMENTED FUNCTION: tcdrain 290381297 Cancellation of pthread_cond_wait does not require mutex 290482872 Using debug info from additional packages (wishlist) 290583025 Support for ioctls FIGETBSZ and FIBMAP 290683340 Support for ioctl HDIO_GET_IDENTITY 290779714 Support for the semtimedop system call. 290877022 Support for ioctls FBIOGET_VSCREENINFO and FBIOGET_FSCREENINFO 290982098 hp2ps ansification (wishlist) 291083573 Valgrind SIGSEGV on execve 291182999 show which cmdline option was erroneous (wishlist) 291283040 make valgrind VPATH and distcheck-clean (wishlist) 291383998 Assertion `newfd > vgPlain_max_fd' failed (see below) 291482722 Unchecked mmap in as_pad leads to mysterious failures later 291578958 memcheck seg faults while running Mozilla 291685416 Arguments with colon (e.g. --logsocket) ignored 2917 2918 2919Additionally there are the following changes, which are not 2920connected to any bug report numbers, AFAICS: 2921 2922* Rearranged address space layout relative to 2.1.1, so that 2923 Valgrind/tools will run out of memory later than currently in many 2924 circumstances. This is good news esp. for Calltree. It should 2925 be possible for client programs to allocate over 800MB of 2926 memory when using memcheck now. 2927 2928* Improved checking when laying out memory. Should hopefully avoid 2929 the random segmentation faults that 2.1.1 sometimes caused. 2930 2931* Support for Fedora Core 2 and SuSE 9.1. Improvements to NPTL 2932 support to the extent that V now works properly on NPTL-only setups. 2933 2934* Renamed the following options: 2935 --logfile-fd --> --log-fd 2936 --logfile --> --log-file 2937 --logsocket --> --log-socket 2938 to be consistent with each other and other options (esp. --input-fd). 2939 2940* Add support for SIOCGMIIPHY, SIOCGMIIREG and SIOCSMIIREG ioctls and 2941 improve the checking of other interface related ioctls. 2942 2943* Fix building with gcc-3.4.1. 2944 2945* Remove limit on number of semaphores supported. 2946 2947* Add support for syscalls: set_tid_address (258), acct (51). 2948 2949* Support instruction "repne movs" -- not official but seems to occur. 2950 2951* Implement an emulated soft limit for file descriptors in addition to 2952 the current reserved area, which effectively acts as a hard limit. The 2953 setrlimit system call now simply updates the emulated limits as best 2954 as possible - the hard limit is not allowed to move at all and just 2955 returns EPERM if you try and change it. This should stop reductions 2956 in the soft limit causing assertions when valgrind tries to allocate 2957 descriptors from the reserved area. 2958 (This actually came from bug #83998). 2959 2960* Major overhaul of Cachegrind implementation. First user-visible change 2961 is that cachegrind.out files are now typically 90% smaller than they 2962 used to be; code annotation times are correspondingly much smaller. 2963 Second user-visible change is that hit/miss counts for code that is 2964 unloaded at run-time is no longer dumped into a single "discard" pile, 2965 but accurately preserved. 2966 2967* Client requests for telling valgrind about memory pools. 2968 2969 2970 2971Developer (cvs head) release 2.1.1 (12 March 2004) 2972~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 29732.1.1 contains some internal structural changes needed for V's 2974long-term future. These don't affect end-users. Most notable 2975user-visible changes are: 2976 2977* Greater isolation between Valgrind and the program being run, so 2978 the program is less likely to inadvertently kill Valgrind by 2979 doing wild writes. 2980 2981* Massif: a new space profiling tool. Try it! It's cool, and it'll 2982 tell you in detail where and when your C/C++ code is allocating heap. 2983 Draws pretty .ps pictures of memory use against time. A potentially 2984 powerful tool for making sense of your program's space use. 2985 2986* Fixes for many bugs, including support for more SSE2/SSE3 instructions, 2987 various signal/syscall things, and various problems with debug 2988 info readers. 2989 2990* Support for glibc-2.3.3 based systems. 2991 2992We are now doing automatic overnight build-and-test runs on a variety 2993of distros. As a result, we believe 2.1.1 builds and runs on: 2994Red Hat 7.2, 7.3, 8.0, 9, Fedora Core 1, SuSE 8.2, SuSE 9. 2995 2996 2997The following bugs, and probably many more, have been fixed. These 2998are listed at http://bugs.kde.org. Reporting a bug for valgrind in 2999the http://bugs.kde.org is much more likely to get you a fix than 3000mailing developers directly, so please continue to keep sending bugs 3001there. 3002 300369616 glibc 2.3.2 w/NPTL is massively different than what valgrind expects 300469856 I don't know how to instrument MMXish stuff (Helgrind) 300573892 valgrind segfaults starting with Objective-C debug info 3006 (fix for S-type stabs) 300773145 Valgrind complains too much about close(<reserved fd>) 300873902 Shadow memory allocation seems to fail on RedHat 8.0 300968633 VG_N_SEMAPHORES too low (V itself was leaking semaphores) 301075099 impossible to trace multiprocess programs 301176839 the `impossible' happened: disInstr: INT but not 0x80 ! 301276762 vg_to_ucode.c:3748 (dis_push_segreg): Assertion `sz == 4' failed. 301376747 cannot include valgrind.h in c++ program 301476223 parsing B(3,10) gave NULL type => impossible happens 301575604 shmdt handling problem 301676416 Problems with gcc 3.4 snap 20040225 301775614 using -gstabs when building your programs the `impossible' happened 301875787 Patch for some CDROM ioctls CDORM_GET_MCN, CDROM_SEND_PACKET, 301975294 gcc 3.4 snapshot's libstdc++ have unsupported instructions. 3020 (REP RET) 302173326 vg_symtab2.c:272 (addScopeRange): Assertion `range->size > 0' failed. 302272596 not recognizing __libc_malloc 302369489 Would like to attach ddd to running program 302472781 Cachegrind crashes with kde programs 302573055 Illegal operand at DXTCV11CompressBlockSSE2 (more SSE opcodes) 302673026 Descriptor leak check reports port numbers wrongly 302771705 README_MISSING_SYSCALL_OR_IOCTL out of date 302872643 Improve support for SSE/SSE2 instructions 302972484 valgrind leaves it's own signal mask in place when execing 303072650 Signal Handling always seems to restart system calls 303172006 The mmap system call turns all errors in ENOMEM 303271781 gdb attach is pretty useless 303371180 unhandled instruction bytes: 0xF 0xAE 0x85 0xE8 303469886 writes to zero page cause valgrind to assert on exit 303571791 crash when valgrinding gimp 1.3 (stabs reader problem) 303669783 unhandled syscall: 218 303769782 unhandled instruction bytes: 0x66 0xF 0x2B 0x80 303870385 valgrind fails if the soft file descriptor limit is less 3039 than about 828 304069529 "rep; nop" should do a yield 304170827 programs with lots of shared libraries report "mmap failed" 3042 for some of them when reading symbols 304371028 glibc's strnlen is optimised enough to confuse valgrind 3044 3045 3046 3047 3048Unstable (cvs head) release 2.1.0 (15 December 2003) 3049~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3050For whatever it's worth, 2.1.0 actually seems pretty darn stable to me 3051(Julian). It looks eminently usable, and given that it fixes some 3052significant bugs, may well be worth using on a day-to-day basis. 30532.1.0 is known to build and pass regression tests on: SuSE 9, SuSE 30548.2, RedHat 8. 3055 30562.1.0 most notably includes Jeremy Fitzhardinge's complete overhaul of 3057handling of system calls and signals, and their interaction with 3058threads. In general, the accuracy of the system call, thread and 3059signal simulations is much improved. Specifically: 3060 3061- Blocking system calls behave exactly as they do when running 3062 natively (not on valgrind). That is, if a syscall blocks only the 3063 calling thread when running natively, than it behaves the same on 3064 valgrind. No more mysterious hangs because V doesn't know that some 3065 syscall or other, should block only the calling thread. 3066 3067- Interrupted syscalls should now give more faithful results. 3068 3069- Finally, signal contexts in signal handlers are supported. As a 3070 result, konqueror on SuSE 9 no longer segfaults when notified of 3071 file changes in directories it is watching. 3072 3073Other changes: 3074 3075- Robert Walsh's file descriptor leakage checks. When enabled, 3076 Valgrind will print out a list of open file descriptors on 3077 exit. Along with each file descriptor, Valgrind prints out a stack 3078 backtrace of where the file was opened and any details relating to the 3079 file descriptor such as the file name or socket details. 3080 To use, give: --track-fds=yes 3081 3082- Implemented a few more SSE/SSE2 instructions. 3083 3084- Less crud on the stack when you do 'where' inside a GDB attach. 3085 3086- Fixed the following bugs: 3087 68360: Valgrind does not compile against 2.6.0-testX kernels 3088 68525: CVS head doesn't compile on C90 compilers 3089 68566: pkgconfig support (wishlist) 3090 68588: Assertion `sz == 4' failed in vg_to_ucode.c (disInstr) 3091 69140: valgrind not able to explicitly specify a path to a binary. 3092 69432: helgrind asserts encountering a MutexErr when there are 3093 EraserErr suppressions 3094 3095- Increase the max size of the translation cache from 200k average bbs 3096 to 300k average bbs. Programs on the size of OOo (680m17) are 3097 thrashing the cache at the smaller size, creating large numbers of 3098 retranslations and wasting significant time as a result. 3099 3100 3101 3102Stable release 2.0.0 (5 Nov 2003) 3103~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3104 31052.0.0 improves SSE/SSE2 support, fixes some minor bugs, and 3106improves support for SuSE 9 and the Red Hat "Severn" beta. 3107 3108- Further improvements to SSE/SSE2 support. The entire test suite of 3109 the GNU Scientific Library (gsl-1.4) compiled with Intel Icc 7.1 3110 20030307Z '-g -O -xW' now works. I think this gives pretty good 3111 coverage of SSE/SSE2 floating point instructions, or at least the 3112 subset emitted by Icc. 3113 3114- Also added support for the following instructions: 3115 MOVNTDQ UCOMISD UNPCKLPS UNPCKHPS SQRTSS 3116 PUSH/POP %{FS,GS}, and PUSH %CS (Nb: there is no POP %CS). 3117 3118- CFI support for GDB version 6. Needed to enable newer GDBs 3119 to figure out where they are when using --gdb-attach=yes. 3120 3121- Fix this: 3122 mc_translate.c:1091 (memcheck_instrument): Assertion 3123 `u_in->size == 4 || u_in->size == 16' failed. 3124 3125- Return an error rather than panicing when given a bad socketcall. 3126 3127- Fix checking of syscall rt_sigtimedwait(). 3128 3129- Implement __NR_clock_gettime (syscall 265). Needed on Red Hat Severn. 3130 3131- Fixed bug in overlap check in strncpy() -- it was assuming the src was 'n' 3132 bytes long, when it could be shorter, which could cause false 3133 positives. 3134 3135- Support use of select() for very large numbers of file descriptors. 3136 3137- Don't fail silently if the executable is statically linked, or is 3138 setuid/setgid. Print an error message instead. 3139 3140- Support for old DWARF-1 format line number info. 3141 3142 3143 3144Snapshot 20031012 (12 October 2003) 3145~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3146 3147Three months worth of bug fixes, roughly. Most significant single 3148change is improved SSE/SSE2 support, mostly thanks to Dirk Mueller. 3149 315020031012 builds on Red Hat Fedora ("Severn") but doesn't really work 3151(curiosly, mozilla runs OK, but a modest "ls -l" bombs). I hope to 3152get a working version out soon. It may or may not work ok on the 3153forthcoming SuSE 9; I hear positive noises about it but haven't been 3154able to verify this myself (not until I get hold of a copy of 9). 3155 3156A detailed list of changes, in no particular order: 3157 3158- Describe --gen-suppressions in the FAQ. 3159 3160- Syscall __NR_waitpid supported. 3161 3162- Minor MMX bug fix. 3163 3164- -v prints program's argv[] at startup. 3165 3166- More glibc-2.3 suppressions. 3167 3168- Suppressions for stack underrun bug(s) in the c++ support library 3169 distributed with Intel Icc 7.0. 3170 3171- Fix problems reading /proc/self/maps. 3172 3173- Fix a couple of messages that should have been suppressed by -q, 3174 but weren't. 3175 3176- Make Addrcheck understand "Overlap" suppressions. 3177 3178- At startup, check if program is statically linked and bail out if so. 3179 3180- Cachegrind: Auto-detect Intel Pentium-M, also VIA Nehemiah 3181 3182- Memcheck/addrcheck: minor speed optimisations 3183 3184- Handle syscall __NR_brk more correctly than before. 3185 3186- Fixed incorrect allocate/free mismatch errors when using 3187 operator new(unsigned, std::nothrow_t const&) 3188 operator new[](unsigned, std::nothrow_t const&) 3189 3190- Support POSIX pthread spinlocks. 3191 3192- Fixups for clean compilation with gcc-3.3.1. 3193 3194- Implemented more opcodes: 3195 - push %es 3196 - push %ds 3197 - pop %es 3198 - pop %ds 3199 - movntq 3200 - sfence 3201 - pshufw 3202 - pavgb 3203 - ucomiss 3204 - enter 3205 - mov imm32, %esp 3206 - all "in" and "out" opcodes 3207 - inc/dec %esp 3208 - A whole bunch of SSE/SSE2 instructions 3209 3210- Memcheck: don't bomb on SSE/SSE2 code. 3211 3212 3213Snapshot 20030725 (25 July 2003) 3214~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3215 3216Fixes some minor problems in 20030716. 3217 3218- Fix bugs in overlap checking for strcpy/memcpy etc. 3219 3220- Do overlap checking with Addrcheck as well as Memcheck. 3221 3222- Fix this: 3223 Memcheck: the `impossible' happened: 3224 get_error_name: unexpected type 3225 3226- Install headers needed to compile new skins. 3227 3228- Remove leading spaces and colon in the LD_LIBRARY_PATH / LD_PRELOAD 3229 passed to non-traced children. 3230 3231- Fix file descriptor leak in valgrind-listener. 3232 3233- Fix longstanding bug in which the allocation point of a 3234 block resized by realloc was not correctly set. This may 3235 have caused confusing error messages. 3236 3237 3238Snapshot 20030716 (16 July 2003) 3239~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3240 324120030716 is a snapshot of our current CVS head (development) branch. 3242This is the branch which will become valgrind-2.0. It contains 3243significant enhancements over the 1.9.X branch. 3244 3245Despite this being a snapshot of the CVS head, it is believed to be 3246quite stable -- at least as stable as 1.9.6 or 1.0.4, if not more so 3247-- and therefore suitable for widespread use. Please let us know asap 3248if it causes problems for you. 3249 3250Two reasons for releasing a snapshot now are: 3251 3252- It's been a while since 1.9.6, and this snapshot fixes 3253 various problems that 1.9.6 has with threaded programs 3254 on glibc-2.3.X based systems. 3255 3256- So as to make available improvements in the 2.0 line. 3257 3258Major changes in 20030716, as compared to 1.9.6: 3259 3260- More fixes to threading support on glibc-2.3.1 and 2.3.2-based 3261 systems (SuSE 8.2, Red Hat 9). If you have had problems 3262 with inconsistent/illogical behaviour of errno, h_errno or the DNS 3263 resolver functions in threaded programs, 20030716 should improve 3264 matters. This snapshot seems stable enough to run OpenOffice.org 3265 1.1rc on Red Hat 7.3, SuSE 8.2 and Red Hat 9, and that's a big 3266 threaded app if ever I saw one. 3267 3268- Automatic generation of suppression records; you no longer 3269 need to write them by hand. Use --gen-suppressions=yes. 3270 3271- strcpy/memcpy/etc check their arguments for overlaps, when 3272 running with the Memcheck or Addrcheck skins. 3273 3274- malloc_usable_size() is now supported. 3275 3276- new client requests: 3277 - VALGRIND_COUNT_ERRORS, VALGRIND_COUNT_LEAKS: 3278 useful with regression testing 3279 - VALGRIND_NON_SIMD_CALL[0123]: for running arbitrary functions 3280 on real CPU (use with caution!) 3281 3282- The GDB attach mechanism is more flexible. Allow the GDB to 3283 be run to be specified by --gdb-path=/path/to/gdb, and specify 3284 which file descriptor V will read its input from with 3285 --input-fd=<number>. 3286 3287- Cachegrind gives more accurate results (wasn't tracking instructions in 3288 malloc() and friends previously, is now). 3289 3290- Complete support for the MMX instruction set. 3291 3292- Partial support for the SSE and SSE2 instruction sets. Work for this 3293 is ongoing. About half the SSE/SSE2 instructions are done, so 3294 some SSE based programs may work. Currently you need to specify 3295 --skin=addrcheck. Basically not suitable for real use yet. 3296 3297- Significant speedups (10%-20%) for standard memory checking. 3298 3299- Fix assertion failure in pthread_once(). 3300 3301- Fix this: 3302 valgrind: vg_intercept.c:598 (vgAllRoadsLeadToRome_select): 3303 Assertion `ms_end >= ms_now' failed. 3304 3305- Implement pthread_mutexattr_setpshared. 3306 3307- Understand Pentium 4 branch hints. Also implemented a couple more 3308 obscure x86 instructions. 3309 3310- Lots of other minor bug fixes. 3311 3312- We have a decent regression test system, for the first time. 3313 This doesn't help you directly, but it does make it a lot easier 3314 for us to track the quality of the system, especially across 3315 multiple linux distributions. 3316 3317 You can run the regression tests with 'make regtest' after 'make 3318 install' completes. On SuSE 8.2 and Red Hat 9 I get this: 3319 3320 == 84 tests, 0 stderr failures, 0 stdout failures == 3321 3322 On Red Hat 8, I get this: 3323 3324 == 84 tests, 2 stderr failures, 1 stdout failure == 3325 corecheck/tests/res_search (stdout) 3326 memcheck/tests/sigaltstack (stderr) 3327 3328 sigaltstack is probably harmless. res_search doesn't work 3329 on R H 8 even running natively, so I'm not too worried. 3330 3331 On Red Hat 7.3, a glibc-2.2.5 system, I get these harmless failures: 3332 3333 == 84 tests, 2 stderr failures, 1 stdout failure == 3334 corecheck/tests/pth_atfork1 (stdout) 3335 corecheck/tests/pth_atfork1 (stderr) 3336 memcheck/tests/sigaltstack (stderr) 3337 3338 You need to run on a PII system, at least, since some tests 3339 contain P6-specific instructions, and the test machine needs 3340 access to the internet so that corecheck/tests/res_search 3341 (a test that the DNS resolver works) can function. 3342 3343As ever, thanks for the vast amount of feedback :) and bug reports :( 3344We may not answer all messages, but we do at least look at all of 3345them, and tend to fix the most frequently reported bugs. 3346 3347 3348 3349Version 1.9.6 (7 May 2003 or thereabouts) 3350~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3351 3352Major changes in 1.9.6: 3353 3354- Improved threading support for glibc >= 2.3.2 (SuSE 8.2, 3355 RedHat 9, to name but two ...) It turned out that 1.9.5 3356 had problems with threading support on glibc >= 2.3.2, 3357 usually manifested by threaded programs deadlocking in system calls, 3358 or running unbelievably slowly. Hopefully these are fixed now. 1.9.6 3359 is the first valgrind which gives reasonable support for 3360 glibc-2.3.2. Also fixed a 2.3.2 problem with pthread_atfork(). 3361 3362- Majorly expanded FAQ.txt. We've added workarounds for all 3363 common problems for which a workaround is known. 3364 3365Minor changes in 1.9.6: 3366 3367- Fix identification of the main thread's stack. Incorrect 3368 identification of it was causing some on-stack addresses to not get 3369 identified as such. This only affected the usefulness of some error 3370 messages; the correctness of the checks made is unchanged. 3371 3372- Support for kernels >= 2.5.68. 3373 3374- Dummy implementations of __libc_current_sigrtmin, 3375 __libc_current_sigrtmax and __libc_allocate_rtsig, hopefully 3376 good enough to keep alive programs which previously died for lack of 3377 them. 3378 3379- Fix bug in the VALGRIND_DISCARD_TRANSLATIONS client request. 3380 3381- Fix bug in the DWARF2 debug line info loader, when instructions 3382 following each other have source lines far from each other 3383 (e.g. with inlined functions). 3384 3385- Debug info reading: read symbols from both "symtab" and "dynsym" 3386 sections, rather than merely from the one that comes last in the 3387 file. 3388 3389- New syscall support: prctl(), creat(), lookup_dcookie(). 3390 3391- When checking calls to accept(), recvfrom(), getsocketopt(), 3392 don't complain if buffer values are NULL. 3393 3394- Try and avoid assertion failures in 3395 mash_LD_PRELOAD_and_LD_LIBRARY_PATH. 3396 3397- Minor bug fixes in cg_annotate. 3398 3399 3400 3401Version 1.9.5 (7 April 2003) 3402~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3403 3404It occurs to me that it would be helpful for valgrind users to record 3405in the source distribution the changes in each release. So I now 3406attempt to mend my errant ways :-) Changes in this and future releases 3407will be documented in the NEWS file in the source distribution. 3408 3409Major changes in 1.9.5: 3410 3411- (Critical bug fix): Fix a bug in the FPU simulation. This was 3412 causing some floating point conditional tests not to work right. 3413 Several people reported this. If you had floating point code which 3414 didn't work right on 1.9.1 to 1.9.4, it's worth trying 1.9.5. 3415 3416- Partial support for Red Hat 9. RH9 uses the new Native Posix 3417 Threads Library (NPTL), instead of the older LinuxThreads. 3418 This potentially causes problems with V which will take some 3419 time to correct. In the meantime we have partially worked around 3420 this, and so 1.9.5 works on RH9. Threaded programs still work, 3421 but they may deadlock, because some system calls (accept, read, 3422 write, etc) which should be nonblocking, in fact do block. This 3423 is a known bug which we are looking into. 3424 3425 If you can, your best bet (unfortunately) is to avoid using 3426 1.9.5 on a Red Hat 9 system, or on any NPTL-based distribution. 3427 If your glibc is 2.3.1 or earlier, you're almost certainly OK. 3428 3429Minor changes in 1.9.5: 3430 3431- Added some #errors to valgrind.h to ensure people don't include 3432 it accidentally in their sources. This is a change from 1.0.X 3433 which was never properly documented. The right thing to include 3434 is now memcheck.h. Some people reported problems and strange 3435 behaviour when (incorrectly) including valgrind.h in code with 3436 1.9.1 -- 1.9.4. This is no longer possible. 3437 3438- Add some __extension__ bits and pieces so that gcc configured 3439 for valgrind-checking compiles even with -Werror. If you 3440 don't understand this, ignore it. Of interest to gcc developers 3441 only. 3442 3443- Removed a pointless check which caused problems interworking 3444 with Clearcase. V would complain about shared objects whose 3445 names did not end ".so", and refuse to run. This is now fixed. 3446 In fact it was fixed in 1.9.4 but not documented. 3447 3448- Fixed a bug causing an assertion failure of "waiters == 1" 3449 somewhere in vg_scheduler.c, when running large threaded apps, 3450 notably MySQL. 3451 3452- Add support for the munlock system call (124). 3453 3454Some comments about future releases: 3455 34561.9.5 is, we hope, the most stable Valgrind so far. It pretty much 3457supersedes the 1.0.X branch. If you are a valgrind packager, please 3458consider making 1.9.5 available to your users. You can regard the 34591.0.X branch as obsolete: 1.9.5 is stable and vastly superior. There 3460are no plans at all for further releases of the 1.0.X branch. 3461 3462If you want a leading-edge valgrind, consider building the cvs head 3463(from SourceForge), or getting a snapshot of it. Current cool stuff 3464going in includes MMX support (done); SSE/SSE2 support (in progress), 3465a significant (10-20%) performance improvement (done), and the usual 3466large collection of minor changes. Hopefully we will be able to 3467improve our NPTL support, but no promises. 3468 3469